Support moe_wna16 as well

Signed-off-by: mgoin <mgoin64@gmail.com>

.buildkite/lm-eval-harness/configs/DeepSeek-V2-Lite-Chat.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash ./run-lm-eval-gsm-vllm-baseline.sh -m deepseek-ai/DeepSeek-V2-Lite-Chat -b "auto" -l 1000 -f 5 -t 2
 model_name: "deepseek-ai/DeepSeek-V2-Lite-Chat"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform.yaml

@@ -1,4 +1,3 @@
-# For hf script, without -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m nm-testing/Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform -b auto -l 1000 -f 5
 model_name: "nm-testing/Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-70B-Instruct.yaml

@@ -1,4 +1,3 @@
-# For hf script, without -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m meta-llama/Meta-Llama-3-70B-Instruct -b 32 -l 250 -f 5
 model_name: "meta-llama/Meta-Llama-3-70B-Instruct"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-Channelwise-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-Instruct-W8A8-FP8-Channelwise-compressed-tensors -b auto -l 1000 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-Instruct-W8A8-FP8-Channelwise-compressed-tensors"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FBGEMM-nonuniform.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-Instruct-FBGEMM-nonuniform -b auto -l 1000 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-Instruct-FBGEMM-nonuniform"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FP8-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test -b 32 -l 1000 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FP8.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m neuralmagic/Meta-Llama-3-8B-Instruct-FP8 -b 32 -l 250 -f 5 -t 1
 model_name: "neuralmagic/Meta-Llama-3-8B-Instruct-FP8"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-INT8-compressed-tensors-asym.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-Instruct-W8-Channel-A8-Dynamic-Asym-Per-Token-Test -b "auto" -l 250 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-Instruct-W8-Channel-A8-Dynamic-Asym-Per-Token-Test"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-INT8-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-Instruct-W8-Channel-A8-Dynamic-Per-Token-Test -b "auto" -l 250 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-Instruct-W8-Channel-A8-Dynamic-Per-Token-Test"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-nonuniform-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-Instruct-nonuniform-test -b auto -l 1000 -f 5 -t 1
 model_name: "nm-testing/Meta-Llama-3-8B-Instruct-nonuniform-test"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct.yaml

@@ -1,5 +1,4 @@
-# For hf script, without -t option (tensor parallel size).
-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m meta-llama/Meta-Llama-3-8B-Instruct -b 32 -l 250 -f 5
+# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m meta-llama/Meta-Llama-3-8B-Instruct -b 32 -l 250 -f 5 -t 1
 model_name: "meta-llama/Meta-Llama-3-8B-Instruct"
 tasks:
 - name: "gsm8k"

.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-QQQ.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m HandH1998/QQQ-Llama-3-8b-g128 -b 32 -l 1000 -f 5 -t 1
 model_name: "HandH1998/QQQ-Llama-3-8b-g128"
 tasks:

.buildkite/lm-eval-harness/configs/Meta-Llama-3.2-1B-Instruct-FP8-compressed-tensors.yaml

@@ -1,11 +0,0 @@
-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m RedHatAI/Llama-3.2-1B-Instruct-FP8 -b "auto" -l 1319 -f 5 -t 1
-model_name: "RedHatAI/Llama-3.2-1B-Instruct-FP8"
-tasks:
-- name: "gsm8k"
-  metrics:
-  - name: "exact_match,strict-match"
-    value: 0.335
-  - name: "exact_match,flexible-extract"
-    value: 0.323
-limit: 1319
-num_fewshot: 5

.buildkite/lm-eval-harness/configs/Meta-Llama-3.2-1B-Instruct-INT8-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m neuralmagic/Llama-3.2-1B-Instruct-quantized.w8a8 -b "auto" -l 1000 -f 5 -t 1
 model_name: "neuralmagic/Llama-3.2-1B-Instruct-quantized.w8a8"
 tasks:

.buildkite/lm-eval-harness/configs/Minitron-4B-Base-FP8.yaml

@@ -1,12 +1,11 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m mgoin/Minitron-4B-Base-FP8 -b auto -l 1000 -f 5 -t 1
 model_name: "mgoin/Minitron-4B-Base-FP8"
 tasks:
 - name: "gsm8k"
   metrics:
   - name: "exact_match,strict-match"
-    value: 0.231
+    value: 0.233
   - name: "exact_match,flexible-extract"
-    value: 0.22
+    value: 0.236
 limit: 1000
 num_fewshot: 5

.buildkite/lm-eval-harness/configs/Mixtral-8x22B-Instruct-v0.1-FP8-Dynamic.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash ./run-lm-eval-gsm-vllm-baseline.sh -m neuralmagic/Mixtral-8x22B-Instruct-v0.1-FP8-dynamic -b "auto" -l 250 -f 5 -t 8
 model_name: "neuralmagic/Mixtral-8x22B-Instruct-v0.1-FP8-dynamic"
 tasks:

.buildkite/lm-eval-harness/configs/Mixtral-8x7B-Instruct-v0.1-FP8.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash ./run-lm-eval-gsm-vllm-baseline.sh -m neuralmagic/Mixtral-8x7B-Instruct-v0.1-FP8 -b "auto" -l 250 -f 5 -t 4
 model_name: "neuralmagic/Mixtral-8x7B-Instruct-v0.1-FP8"
 tasks:

.buildkite/lm-eval-harness/configs/Mixtral-8x7B-Instruct-v0.1.yaml

@@ -1,5 +1,4 @@
-# For hf script, without -t option (tensor parallel size).
-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m neuralmagic/Mixtral-8x7B-Instruct-v0.1 -b 32 -l 250 -f 5
+# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m neuralmagic/Mixtral-8x7B-Instruct-v0.1 -b 32 -l 250 -f 5 -t 4
 model_name: "mistralai/Mixtral-8x7B-Instruct-v0.1"
 tasks:
 - name: "gsm8k"

.buildkite/lm-eval-harness/configs/Qwen1.5-MoE-W4A16-compressed-tensors.yaml

@@ -1,12 +0,0 @@
-# For vllm script, with -t option (tensor parallel size).
-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Qwen1.5-MoE-A2.7B-Chat-quantized.w4a16 -b auto -l 1319 -f 5 -t 1
-model_name: "nm-testing/Qwen1.5-MoE-A2.7B-Chat-quantized.w4a16"
-tasks:
-- name: "gsm8k"
-  metrics:
-  - name: "exact_match,strict-match"
-    value: 0.30
-  - name: "exact_match,flexible-extract"
-    value: 0.465
-limit: 1319
-num_fewshot: 5

.buildkite/lm-eval-harness/configs/Qwen2-1.5B-Instruct-FP8W8.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Qwen2-1.5B-Instruct-FP8W8 -b auto -l 1000 -f 5 -t 1
 model_name: "nm-testing/Qwen2-1.5B-Instruct-FP8W8"
 tasks:

.buildkite/lm-eval-harness/configs/Qwen2-1.5B-Instruct-INT8-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m neuralmagic/Qwen2-1.5B-Instruct-quantized.w8a8 -b "auto" -l 1000 -f 5 -t 1
 model_name: "neuralmagic/Qwen2-1.5B-Instruct-quantized.w8a8"
 tasks:

.buildkite/lm-eval-harness/configs/Qwen2-1.5B-Instruct-W8A16-compressed-tensors.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Qwen2-1.5B-Instruct-W8A16-Channelwise -b "auto" -l 1000 -f 5 -t 1
 model_name: "nm-testing/Qwen2-1.5B-Instruct-W8A16-Channelwise"
 tasks:

.buildkite/lm-eval-harness/configs/Qwen2-57B-A14-Instruct.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash ./run-lm-eval-gsm-vllm-baseline.sh -m Qwen/Qwen2-57B-A14B-Instruct -b "auto" -l 250 -f 5 -t 4
 model_name: "Qwen/Qwen2-57B-A14B-Instruct"
 tasks:

.buildkite/lm-eval-harness/configs/Qwen2.5-1.5B-Instruct.yaml

@@ -1,11 +0,0 @@
-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m Qwen/Qwen2.5-1.5B-Instruct -b auto -l 1319 -f 5 -t 1
-model_name: "Qwen/Qwen2.5-1.5B-Instruct"
-tasks:
-- name: "gsm8k"
-  metrics:
-  - name: "exact_match,strict-match"
-    value: 0.54
-  - name: "exact_match,flexible-extract"
-    value: 0.59
-limit: 1319
-num_fewshot: 5

.buildkite/lm-eval-harness/configs/Qwen2.5-VL-3B-Instruct-FP8-dynamic.yaml

@@ -1,11 +0,0 @@
-# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m RedHatAI/Qwen2.5-VL-3B-Instruct-FP8-Dynamic -b auto -l 1319 -f 5 -t 1
-model_name: "RedHatAI/Qwen2.5-VL-3B-Instruct-FP8-Dynamic"
-tasks:
-- name: "gsm8k"
-  metrics:
-  - name: "exact_match,strict-match"
-    value: 0.47
-  - name: "exact_match,flexible-extract"
-    value: 0.64
-limit: 1319
-num_fewshot: 5

.buildkite/lm-eval-harness/configs/SparseLlama3.1_2of4_fp8_compressed.yaml

@@ -1,4 +1,3 @@
-# For vllm script, with -t option (tensor parallel size).
 # bash ./run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/SparseLlama-3.1-8B-gsm8k-pruned.2of4-chnl_wts_per_tok_dyn_act_fp8-BitM -b "auto" -t 2
 model_name: "nm-testing/SparseLlama-3.1-8B-gsm8k-pruned.2of4-chnl_wts_per_tok_dyn_act_fp8-BitM"
 tasks:

.buildkite/lm-eval-harness/configs/models-large.txt

@@ -3,4 +3,3 @@ Meta-Llama-3-70B-Instruct.yaml
 Mixtral-8x7B-Instruct-v0.1.yaml
 Qwen2-57B-A14-Instruct.yaml
 DeepSeek-V2-Lite-Chat.yaml
-Meta-Llama-3-8B-QQQ.yaml

.buildkite/lm-eval-harness/configs/models-small.txt

@@ -1,6 +1,10 @@
-Qwen2.5-1.5B-Instruct.yaml
+Meta-Llama-3-8B-Instruct.yaml
+Meta-Llama-3-8B-Instruct-FP8-compressed-tensors.yaml
 Meta-Llama-3.2-1B-Instruct-INT8-compressed-tensors.yaml
 Meta-Llama-3-8B-Instruct-INT8-compressed-tensors-asym.yaml
 Meta-Llama-3-8B-Instruct-nonuniform-compressed-tensors.yaml
-Qwen2.5-VL-3B-Instruct-FP8-dynamic.yaml
-Qwen1.5-MoE-W4A16-compressed-tensors.yaml
+Meta-Llama-3-8B-Instruct-Channelwise-compressed-tensors.yaml
+Minitron-4B-Base-FP8.yaml
+Qwen2-1.5B-Instruct-INT8-compressed-tensors.yaml
+Qwen2-1.5B-Instruct-FP8W8.yaml
+Meta-Llama-3-8B-QQQ.yaml

.buildkite/lm-eval-harness/conftest.py

@@ -1,39 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-from pathlib import Path
-
-import pytest
-
-
-def pytest_addoption(parser):
-    parser.addoption(
-        "--config-list-file",
-        action="store",
-        help="Path to the file listing model config YAMLs (one per line)")
-    parser.addoption("--tp-size",
-                     action="store",
-                     default="1",
-                     help="Tensor parallel size to use for evaluation")
-
-
-@pytest.fixture(scope="session")
-def config_list_file(pytestconfig, config_dir):
-    rel_path = pytestconfig.getoption("--config-list-file")
-    return config_dir / rel_path
-
-
-@pytest.fixture(scope="session")
-def tp_size(pytestconfig):
-    return pytestconfig.getoption("--tp-size")
-
-
-def pytest_generate_tests(metafunc):
-    if "config_filename" in metafunc.fixturenames:
-        rel_path = metafunc.config.getoption("--config-list-file")
-        config_list_file = Path(rel_path).resolve()
-        config_dir = config_list_file.parent
-        with open(config_list_file, encoding="utf-8") as f:
-            configs = [
-                config_dir / line.strip() for line in f
-                if line.strip() and not line.startswith("#")
-            ]
-        metafunc.parametrize("config_filename", configs)

.buildkite/lm-eval-harness/run-tests.sh

@@ -0,0 +1,59 @@
+#!/bin/bash
+
+usage() {
+    echo``
+    echo "Runs lm eval harness on GSM8k using vllm and compares to "
+    echo "precomputed baseline (measured by HF transformers.)"
+    echo
+    echo "usage: ${0} <options>"
+    echo
+    echo "  -c    - path to the test data config (e.g. configs/small-models.txt)"
+    echo "  -t    - tensor parallel size"
+    echo
+}
+
+SUCCESS=0
+
+while getopts "c:t:" OPT; do
+  case ${OPT} in
+    c ) 
+        CONFIG="$OPTARG"
+        ;;
+    t )
+        TP_SIZE="$OPTARG"
+        ;;
+    \? )
+        usage
+        exit 1
+        ;;
+  esac
+done
+
+# Parse list of configs.
+IFS=$'\n' read -d '' -r -a MODEL_CONFIGS < "$CONFIG"
+
+for MODEL_CONFIG in "${MODEL_CONFIGS[@]}"
+do
+    LOCAL_SUCCESS=0
+    
+    echo "=== RUNNING MODEL: $MODEL_CONFIG WITH TP SIZE: $TP_SIZE==="
+
+    export LM_EVAL_TEST_DATA_FILE=$PWD/configs/${MODEL_CONFIG}
+    export LM_EVAL_TP_SIZE=$TP_SIZE
+    pytest -s test_lm_eval_correctness.py || LOCAL_SUCCESS=$?
+
+    if [[ $LOCAL_SUCCESS == 0 ]]; then
+        echo "=== PASSED MODEL: ${MODEL_CONFIG} ==="
+    else
+        echo "=== FAILED MODEL: ${MODEL_CONFIG} ==="
+    fi
+
+    SUCCESS=$((SUCCESS + LOCAL_SUCCESS))
+
+done
+
+if [ "${SUCCESS}" -eq "0" ]; then
+    exit 0
+else
+    exit 1
+fi

.buildkite/lm-eval-harness/test_lm_eval_correctness.py

@@ -3,25 +3,34 @@
 LM eval harness on model to compare vs HF baseline computed offline.
 Configs are found in configs/$MODEL.yaml
 
-pytest -s -v test_lm_eval_correctness.py \
-    --config-list-file=configs/models-small.txt \
-    --tp-size=1
+* export LM_EVAL_TEST_DATA_FILE=configs/Meta-Llama-3-70B-Instruct.yaml
+* export LM_EVAL_TP_SIZE=4 
+* pytest -s test_lm_eval_correctness.py
 """
 
+import os
+from pathlib import Path
+
 import lm_eval
-import numpy as np
+import numpy
 import yaml
 
-RTOL = 0.08
+RTOL = 0.05
+TEST_DATA_FILE = os.environ.get(
+    "LM_EVAL_TEST_DATA_FILE",
+    ".buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct.yaml")
+
+TP_SIZE = os.environ.get("LM_EVAL_TP_SIZE", 1)
 
 
-def launch_lm_eval(eval_config, tp_size):
+def launch_lm_eval(eval_config):
     trust_remote_code = eval_config.get('trust_remote_code', False)
+
     model_args = f"pretrained={eval_config['model_name']}," \
-                 f"tensor_parallel_size={tp_size}," \
-                 f"enforce_eager=true," \
+                 f"tensor_parallel_size={TP_SIZE}," \
                  f"add_bos_token=true," \
                  f"trust_remote_code={trust_remote_code}"
+
     results = lm_eval.simple_evaluate(
         model="vllm",
         model_args=model_args,
@@ -29,14 +38,18 @@ def launch_lm_eval(eval_config, tp_size):
         num_fewshot=eval_config["num_fewshot"],
         limit=eval_config["limit"],
         batch_size="auto")
+
     return results
 
 
-def test_lm_eval_correctness_param(config_filename, tp_size):
-    eval_config = yaml.safe_load(config_filename.read_text(encoding="utf-8"))
+def test_lm_eval_correctness():
+    eval_config = yaml.safe_load(
+        Path(TEST_DATA_FILE).read_text(encoding="utf-8"))
 
-    results = launch_lm_eval(eval_config, tp_size)
+    # Launch eval requests.
+    results = launch_lm_eval(eval_config)
 
+    # Confirm scores match ground truth.
     success = True
     for task in eval_config["tasks"]:
         for metric in task["metrics"]:
@@ -44,7 +57,8 @@ def test_lm_eval_correctness_param(config_filename, tp_size):
             measured_value = results["results"][task["name"]][metric["name"]]
             print(f'{task["name"]} | {metric["name"]}: '
                   f'ground_truth={ground_truth} | measured={measured_value}')
-            success = success and np.isclose(
+            success = success and numpy.isclose(
                 ground_truth, measured_value, rtol=RTOL)
 
+    # Assert at the end, print all scores even on failure for debugging.
     assert success

.buildkite/nightly-benchmarks/benchmark-pipeline.yaml

@@ -10,18 +10,12 @@ steps:
           - image: badouralix/curl-jq
             command:
             - sh .buildkite/nightly-benchmarks/scripts/wait-for-image.sh
-  - label: "Cleanup H100"
-    agents:
-      queue: H100
-    depends_on: ~
-    command: docker system prune -a --volumes --force
-  
+
   - label: "A100"
     # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
     agents:
       queue: A100
     depends_on: wait-for-container-image
-    if: build.branch == "main"
     plugins:
     - kubernetes:
         podSpec:
@@ -56,7 +50,6 @@ steps:
     agents:
       queue: H200
     depends_on: wait-for-container-image
-    if: build.branch == "main"
     plugins:
     - docker#v5.12.0:
         image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
@@ -82,7 +75,6 @@ steps:
     agents:
       queue: H100
     depends_on: wait-for-container-image
-    if: build.branch == "main"
     plugins:
     - docker#v5.12.0:
         image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
@@ -98,87 +90,3 @@ steps:
         environment:
         - VLLM_USAGE_SOURCE
         - HF_TOKEN
-
-  # Premerge benchmark
-  - label: "A100"
-    # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
-    agents:
-      queue: A100
-    depends_on: wait-for-container-image
-    if: build.branch != "main"
-    plugins:
-    - kubernetes:
-        podSpec:
-          priorityClassName: perf-benchmark
-          containers:
-          - image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-            command:
-            - bash .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
-            resources:
-              limits:
-                nvidia.com/gpu: 8
-            volumeMounts:
-            - name: devshm
-              mountPath: /dev/shm
-            env:
-            - name: VLLM_USAGE_SOURCE
-              value: ci-test
-            - name: HF_TOKEN
-              valueFrom:
-                secretKeyRef:
-                  name: hf-token-secret
-                  key: token
-          nodeSelector:
-            nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
-          volumes:
-          - name: devshm
-            emptyDir:
-              medium: Memory
-
-  - label: "H200"
-    # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
-    agents:
-      queue: H200
-    depends_on: wait-for-container-image
-    if: build.branch != "main"
-    plugins:
-    - docker#v5.12.0:
-        image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-        command:
-        - bash
-        - .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
-        mount-buildkite-agent: true
-        propagate-environment: true
-        ipc: host
-        gpus: 4,5,6,7
-        volumes:
-          - /data/benchmark-hf-cache:/root/.cache/huggingface
-        environment:
-        - VLLM_USAGE_SOURCE
-        - HF_TOKEN
-
-  #- block: "Run H100 Benchmark"
-    #key: block-h100
-    #depends_on: ~
-
-  - label: "H100"
-    # skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
-    agents:
-      queue: H100
-    depends_on: wait-for-container-image
-    if: build.branch != "main"
-    plugins:
-    - docker#v5.12.0:
-        image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-        command:
-        - bash
-        - .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
-        mount-buildkite-agent: true
-        propagate-environment: true
-        ipc: host
-        gpus: all # see CUDA_VISIBLE_DEVICES for actual GPUs used
-        volumes:
-          - /data/benchmark-hf-cache:/root/.cache/huggingface
-        environment:
-        - VLLM_USAGE_SOURCE
-        - HF_TOKEN

.buildkite/nightly-benchmarks/scripts/convert-results-json-to-markdown.py

@@ -84,13 +84,8 @@ if __name__ == "__main__":
             # this result is generated via `benchmark_serving.py`
 
             # attach the benchmarking command to raw_result
-            try:
-                with open(test_file.with_suffix(".commands")) as f:
-                    command = json.loads(f.read())
-            except OSError as e:
-                print(e)
-                continue
-
+            with open(test_file.with_suffix(".commands")) as f:
+                command = json.loads(f.read())
             raw_result.update(command)
 
             # update the test name of this result
@@ -104,13 +99,8 @@ if __name__ == "__main__":
             # this result is generated via `benchmark_latency.py`
 
             # attach the benchmarking command to raw_result
-            try:
-                with open(test_file.with_suffix(".commands")) as f:
-                    command = json.loads(f.read())
-            except OSError as e:
-                print(e)
-                continue
-
+            with open(test_file.with_suffix(".commands")) as f:
+                command = json.loads(f.read())
             raw_result.update(command)
 
             # update the test name of this result
@@ -131,13 +121,8 @@ if __name__ == "__main__":
             # this result is generated via `benchmark_throughput.py`
 
             # attach the benchmarking command to raw_result
-            try:
-                with open(test_file.with_suffix(".commands")) as f:
-                    command = json.loads(f.read())
-            except OSError as e:
-                print(e)
-                continue
-
+            with open(test_file.with_suffix(".commands")) as f:
+                command = json.loads(f.read())
             raw_result.update(command)
 
             # update the test name of this result

.buildkite/nightly-benchmarks/scripts/run-nightly-benchmarks.sh

@@ -426,7 +426,7 @@ main() {
 
   pip install -U transformers
 
-  pip install -r requirements/dev.txt
+  pip install -r requirements-dev.txt
   which genai-perf
 
   # check storage

.buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh

@@ -10,24 +10,15 @@ set -x
 set -o pipefail
 
 check_gpus() {
-  if command -v nvidia-smi; then
-    # check the number of GPUs and GPU type.
-    declare -g gpu_count=$(nvidia-smi --list-gpus | wc -l)
-  elif command -v amd-smi; then
-    declare -g gpu_count=$(amd-smi list | grep 'GPU' | wc -l)
-  fi
-
+  # check the number of GPUs and GPU type.
+  declare -g gpu_count=$(nvidia-smi --list-gpus | wc -l)
   if [[ $gpu_count -gt 0 ]]; then
     echo "GPU found."
   else
     echo "Need at least 1 GPU to run benchmarking."
     exit 1
   fi
-  if command -v nvidia-smi; then
-    declare -g gpu_type=$(nvidia-smi --query-gpu=name --format=csv,noheader | awk '{print $2}')
-  elif command -v amd-smi; then
-    declare -g gpu_type=$(amd-smi static -g 0 -a | grep 'MARKET_NAME' | awk '{print $2}')
-  fi
+  declare -g gpu_type=$(nvidia-smi --query-gpu=name --format=csv,noheader | awk '{print $2}')
   echo "GPU type is $gpu_type"
 }
 
@@ -99,15 +90,9 @@ kill_gpu_processes() {
 
 
   # wait until GPU memory usage smaller than 1GB
-  if command -v nvidia-smi; then
-    while [ "$(nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits | head -n 1)" -ge 1000 ]; do
-      sleep 1
-    done
-  elif command -v amd-smi; then
-    while [ "$(amd-smi metric -g 0 | grep 'USED_VRAM' | awk '{print $2}')" -ge 1000 ]; do
-      sleep 1
-    done
-  fi
+  while [ "$(nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits | head -n 1)" -ge 1000 ]; do
+    sleep 1
+  done
 
   # remove vllm config file
   rm -rf ~/.config/vllm
@@ -324,14 +309,11 @@ run_serving_tests() {
 
       new_test_name=$test_name"_qps_"$qps
 
-      # pass the tensor parallel size to the client so that it can be displayed
-      # on the benchmark dashboard
       client_command="python3 benchmark_serving.py \
         --save-result \
         --result-dir $RESULTS_FOLDER \
         --result-filename ${new_test_name}.json \
         --request-rate $qps \
-        --metadata "tensor_parallel_size=$tp" \
         $client_args"
 
       echo "Running test case $test_name with qps $qps"
@@ -363,11 +345,6 @@ main() {
   check_gpus
   check_hf_token
 
-  # Set to v1 to run v1 benchmark
-  if [[ "${ENGINE_VERSION:-v0}" == "v1" ]]; then
-    export VLLM_USE_V1=1
-  fi
-
   # dependencies
   (which wget && which curl) || (apt-get update && apt-get install -y wget curl)
   (which jq) || (apt-get update && apt-get -y install jq)
@@ -376,7 +353,7 @@ main() {
   # get the current IP address, required by benchmark_serving.py
   export VLLM_HOST_IP=$(hostname -I | awk '{print $1}')
   # turn of the reporting of the status of each request, to clean up the terminal output
-  export VLLM_LOGGING_LEVEL="WARNING"
+  export VLLM_LOG_LEVEL="WARNING"
 
   # prepare for benchmarking
   cd benchmarks || exit 1

.buildkite/nightly-benchmarks/scripts/wait-for-image.sh

@@ -1,10 +1,6 @@
 #!/bin/sh
 TOKEN=$(curl -s -L "https://public.ecr.aws/token?service=public.ecr.aws&scope=repository:q9t5s3a7/vllm-ci-postmerge-repo:pull" | jq -r .token)
-if [[ "$BUILDKITE_BRANCH" == "main" ]]; then
-    URL="https://public.ecr.aws/v2/q9t5s3a7/vllm-ci-postmerge-repo/manifests/$BUILDKITE_COMMIT"
-else
-    URL="https://public.ecr.aws/v2/q9t5s3a7/vllm-ci-test-repo/manifests/$BUILDKITE_COMMIT"
-fi
+URL="https://public.ecr.aws/v2/q9t5s3a7/vllm-ci-postmerge-repo/manifests/$BUILDKITE_COMMIT"
 
 TIMEOUT_SECONDS=10
 

.buildkite/nightly-benchmarks/tests/latency-tests.json

@@ -29,4 +29,4 @@
             "num-iters": 15
         }
     }
-]
+]

.buildkite/nightly-benchmarks/tests/serving-tests.json

@@ -63,12 +63,11 @@
             "model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
             "disable_log_requests": "", 
             "tensor_parallel_size": 4,
-            "swap_space": 16,
-            "speculative_config": {
-                "model": "turboderp/Qwama-0.5B-Instruct",
-                "num_speculative_tokens": 4,
-                "draft_tensor_parallel_size": 1
-            }
+            "swap_space": 16, 
+            "speculative_model": "turboderp/Qwama-0.5B-Instruct",
+            "num_speculative_tokens": 4,
+            "speculative_draft_tensor_parallel_size": 1,
+            "use_v2_block_manager": ""
         },
         "client_parameters": {
             "model": "meta-llama/Meta-Llama-3.1-70B-Instruct",

.buildkite/nightly-benchmarks/tests/throughput-tests.json

@@ -32,4 +32,4 @@
             "backend": "vllm"
         }
     }
-]
+]

.buildkite/release-pipeline.yaml

@@ -1,23 +1,12 @@
 steps:
-  - label: "Build wheel - CUDA 12.8"
+  - label: "Build wheel - CUDA 12.1"
     agents:
       queue: cpu_queue_postmerge
     commands:
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.8.1 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.1.0 --tag vllm-ci:build-image --target build --progress plain ."
       - "mkdir artifacts"
       - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh"
-    env:
-      DOCKER_BUILDKIT: "1"
-
-  - label: "Build wheel - CUDA 12.6"
-    agents:
-      queue: cpu_queue_postmerge
-    commands:
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.6.3 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
-      - "mkdir artifacts"
-      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh"
+      - "bash .buildkite/upload-wheels.sh"
     env:
       DOCKER_BUILDKIT: "1"
 
@@ -31,10 +20,10 @@ steps:
     agents:
       queue: cpu_queue_postmerge
     commands:
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=11.8.0 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=11.8.0 --tag vllm-ci:build-image --target build --progress plain ."
       - "mkdir artifacts"
       - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh"
+      - "bash .buildkite/upload-wheels.sh"
     env:
       DOCKER_BUILDKIT: "1"
 
@@ -48,7 +37,7 @@ steps:
       queue: cpu_queue_postmerge
     commands:
       - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.8.1 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT --target vllm-openai --progress plain -f docker/Dockerfile ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.1.0 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT --target vllm-openai --progress plain ."
       - "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT"
 
   - label: "Build and publish TPU release image"
@@ -57,9 +46,7 @@ steps:
     agents:
       queue: tpu_queue_postmerge
     commands:
-      - "yes | docker system prune -a"
-      - "git fetch --all"
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --tag vllm/vllm-tpu:nightly --tag vllm/vllm-tpu:$BUILDKITE_COMMIT --progress plain -f docker/Dockerfile.tpu ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --tag vllm/vllm-tpu:nightly --tag vllm/vllm-tpu:$BUILDKITE_COMMIT --progress plain -f Dockerfile.tpu ."
       - "docker push vllm/vllm-tpu:nightly"
       - "docker push vllm/vllm-tpu:$BUILDKITE_COMMIT"
     plugins:
@@ -84,22 +71,7 @@ steps:
       queue: cpu_queue_postmerge
     commands:
       - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:$(buildkite-agent meta-data get release-version) --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:latest --progress plain --target vllm-openai -f docker/Dockerfile.cpu ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:$(buildkite-agent meta-data get release-version) --progress plain -f Dockerfile.cpu ."
       - "docker push public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:$(buildkite-agent meta-data get release-version)"
     env:
       DOCKER_BUILDKIT: "1"
-
-  - block: "Build Neuron release image"
-    key: block-neuron-release-image-build
-    depends_on: ~
-
-  - label: "Build and publish Neuron release image"
-    depends_on: block-neuron-release-image-build
-    agents:
-      queue: neuron-postmerge
-    commands:
-      - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --tag public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:$(buildkite-agent meta-data get release-version) --tag public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:latest --progress plain -f docker/Dockerfile.neuron ."
-      - "docker push public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:$(buildkite-agent meta-data get release-version)"
-    env:
-      DOCKER_BUILDKIT: "1"

.buildkite/run-amd-test.sh

@@ -75,91 +75,36 @@ HF_MOUNT="/root/.cache/huggingface"
 commands=$@
 echo "Commands:$commands"
 #ignore certain kernels tests
-if [[ $commands == *" kernels/core"* ]]; then
+if [[ $commands == *" kernels "* ]]; then
   commands="${commands} \
-  --ignore=kernels/core/test_fused_quant_layernorm.py \
-  --ignore=kernels/core/test_permute_cols.py"
+  --ignore=kernels/test_attention.py \
+  --ignore=kernels/test_attention_selector.py \
+  --ignore=kernels/test_blocksparse_attention.py \
+  --ignore=kernels/test_causal_conv1d.py \
+  --ignore=kernels/test_cutlass.py \
+  --ignore=kernels/test_encoder_decoder_attn.py \
+  --ignore=kernels/test_flash_attn.py \
+  --ignore=kernels/test_flashinfer.py \
+  --ignore=kernels/test_int8_quant.py \
+  --ignore=kernels/test_machete_gemm.py \
+  --ignore=kernels/test_mamba_ssm.py \
+  --ignore=kernels/test_marlin_gemm.py \
+  --ignore=kernels/test_moe.py \
+  --ignore=kernels/test_prefix_prefill.py \
+  --ignore=kernels/test_rand.py \
+  --ignore=kernels/test_sampler.py"
 fi
 
-if [[ $commands == *" kernels/attention"* ]]; then
-  commands="${commands} \
-  --ignore=kernels/attention/stest_attention_selector.py \
-  --ignore=kernels/attention/test_blocksparse_attention.py \
-  --ignore=kernels/attention/test_encoder_decoder_attn.py \
-  --ignore=kernels/attention/test_attention_selector.py \
-  --ignore=kernels/attention/test_flash_attn.py \
-  --ignore=kernels/attention/test_flashinfer.py \
-  --ignore=kernels/attention/test_prefix_prefill.py \
-  --ignore=kernels/attention/test_cascade_flash_attn.py \
-  --ignore=kernels/attention/test_mha_attn.py \
-  --ignore=kernels/attention/test_lightning_attn.py \
-  --ignore=kernels/attention/test_attention.py"
-fi
-
-if [[ $commands == *" kernels/quantization"* ]]; then
-  commands="${commands} \
-  --ignore=kernels/quantization/test_int8_quant.py \
-  --ignore=kernels/quantization/test_aqlm.py \
-  --ignore=kernels/quantization/test_machete_mm.py \
-  --ignore=kernels/quantization/test_block_fp8.py \
-  --ignore=kernels/quantization/test_block_int8.py \
-  --ignore=kernels/quantization/test_marlin_gemm.py \
-  --ignore=kernels/quantization/test_cutlass_scaled_mm.py \
-  --ignore=kernels/quantization/test_int8_kernel.py"
-fi
-
-if [[ $commands == *" kernels/mamba"* ]]; then
-  commands="${commands} \
-  --ignore=kernels/mamba/test_mamba_mixer2.py \
-  --ignore=kernels/mamba/test_causal_conv1d.py \
-  --ignore=kernels/mamba/test_mamba_ssm_ssd.py"
-fi
-
-if [[ $commands == *" kernels/moe"* ]]; then
-  commands="${commands} \
-  --ignore=kernels/moe/test_moe.py \
-  --ignore=kernels/moe/test_cutlass_moe.py \
-  --ignore=kernels/moe/test_triton_moe_ptpc_fp8.py"
-fi
-
-#ignore certain Entrypoints/openai tests
+#ignore certain Entrypoints tests
 if [[ $commands == *" entrypoints/openai "* ]]; then
   commands=${commands//" entrypoints/openai "/" entrypoints/openai \
+  --ignore=entrypoints/openai/test_accuracy.py \
   --ignore=entrypoints/openai/test_audio.py \
-  --ignore=entrypoints/openai/test_shutdown.py \
-  --ignore=entrypoints/openai/test_completion.py \
-  --ignore=entrypoints/openai/test_sleep.py \
-  --ignore=entrypoints/openai/test_models.py \
-  --ignore=entrypoints/openai/test_lora_adapters.py \
-  --ignore=entrypoints/openai/test_return_tokens_as_ids.py \
-  --ignore=entrypoints/openai/test_root_path.py \
-  --ignore=entrypoints/openai/test_tokenization.py \
-  --ignore=entrypoints/openai/test_prompt_validation.py "}
+  --ignore=entrypoints/openai/test_encoder_decoder.py \
+  --ignore=entrypoints/openai/test_embedding.py \
+  --ignore=entrypoints/openai/test_oot_registration.py "}
 fi
 
-#ignore certain Entrypoints/llm tests
-if [[ $commands == *" entrypoints/llm "* ]]; then
-  commands=${commands//" entrypoints/llm "/" entrypoints/llm \
-  --ignore=entrypoints/llm/test_chat.py \
-  --ignore=entrypoints/llm/test_accuracy.py \
-  --ignore=entrypoints/llm/test_init.py \
-  --ignore=entrypoints/llm/test_generate_multiple_loras.py \
-  --ignore=entrypoints/llm/test_prompt_validation.py "}
-fi
-
-#Obsolete currently
-##ignore certain Entrypoints/llm tests
-#if [[ $commands == *" && pytest -v -s entrypoints/llm/test_guided_generate.py"* ]]; then
-#  commands=${commands//" && pytest -v -s entrypoints/llm/test_guided_generate.py"/" "}
-#fi
-
-# --ignore=entrypoints/openai/test_encoder_decoder.py \
-# --ignore=entrypoints/openai/test_embedding.py \
-# --ignore=entrypoints/openai/test_oot_registration.py
-# --ignore=entrypoints/openai/test_accuracy.py \
-# --ignore=entrypoints/openai/test_models.py <= Fails on MI250 but passes on MI300 as of 2025-03-13
-
-
 PARALLEL_JOB_COUNT=8
 # check if the command contains shard flag, we will run all shards in parallel because the host have 8 GPUs. 
 if [[ $commands == *"--shard-id="* ]]; then
@@ -169,16 +114,13 @@ if [[ $commands == *"--shard-id="* ]]; then
     # assign shard-id for each shard
     commands_gpu=${commands//"--shard-id= "/"--shard-id=${GPU} "}
     echo "Shard ${GPU} commands:$commands_gpu"
-    echo "Render devices: $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES"
     docker run \
-        --device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
-        --network=host \
+        --device /dev/kfd --device /dev/dri \
+        --network host \
         --shm-size=16gb \
         --rm \
         -e HIP_VISIBLE_DEVICES="${GPU}" \
         -e HF_TOKEN \
-        -e AWS_ACCESS_KEY_ID \
-        -e AWS_SECRET_ACCESS_KEY \
         -v "${HF_CACHE}:${HF_MOUNT}" \
         -e "HF_HOME=${HF_MOUNT}" \
         --name "${container_name}_${GPU}" \
@@ -199,16 +141,13 @@ if [[ $commands == *"--shard-id="* ]]; then
     fi
   done
 else
-  echo "Render devices: $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES"
   docker run \
-          --device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
-          --network=host \
+          --device /dev/kfd --device /dev/dri \
+          --network host \
           --shm-size=16gb \
           --rm \
           -e HIP_VISIBLE_DEVICES=0 \
           -e HF_TOKEN \
-          -e AWS_ACCESS_KEY_ID \
-          -e AWS_SECRET_ACCESS_KEY \
           -v "${HF_CACHE}:${HF_MOUNT}" \
           -e "HF_HOME=${HF_MOUNT}" \
           --name "${container_name}" \

.buildkite/run-benchmarks.sh

@@ -5,8 +5,8 @@
 set -ex
 set -o pipefail
 
-# cd 2 levels into the working directory
-cd "$(dirname "${BASH_SOURCE[0]}")/../.."
+# cd into parent directory of this file
+cd "$(dirname "${BASH_SOURCE[0]}")/.."
 
 (which wget && which curl) || (apt-get update && apt-get install -y wget curl)
 

.buildkite/run-cpu-test-ppc64le.sh

@@ -10,4 +10,5 @@ trap remove_docker_container EXIT
 remove_docker_container
 
 # Try building the docker image
-docker build -t cpu-test -f docker/Dockerfile.s390x .
+docker build -t cpu-test -f Dockerfile.ppc64le .
+

.buildkite/run-cpu-test.sh

@@ -8,40 +8,34 @@ set -ex
 CORE_RANGE=${CORE_RANGE:-48-95}
 NUMA_NODE=${NUMA_NODE:-1}
 
+# Try building the docker image
+numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build -t cpu-test-"$BUILDKITE_BUILD_NUMBER" -f Dockerfile.cpu .
+numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --build-arg VLLM_CPU_DISABLE_AVX512="true" -t cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2 -f Dockerfile.cpu .
+
 # Setup cleanup
-remove_docker_container() { 
-    set -e; 
-    docker rm -f cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" || true; 
-    docker image rm cpu-test-"$BUILDKITE_BUILD_NUMBER" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2 || true; 
-}
+remove_docker_container() { set -e; docker rm -f cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" || true; }
 trap remove_docker_container EXIT
 remove_docker_container
 
-# Try building the docker image
-numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --tag cpu-test-"$BUILDKITE_BUILD_NUMBER" --target vllm-test -f docker/Dockerfile.cpu .
-numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --build-arg VLLM_CPU_DISABLE_AVX512="true" --tag cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2 --target vllm-test -f docker/Dockerfile.cpu .
-
 # Run the image, setting --shm-size=4g for tensor parallel.
 docker run -itd --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus="$CORE_RANGE"  \
- --cpuset-mems="$NUMA_NODE" --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"
+ --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"
 docker run -itd --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus="$CORE_RANGE" \
- --cpuset-mems="$NUMA_NODE" --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2
+ --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2
 
 function cpu_tests() {
   set -e
   export NUMA_NODE=$2
-  export BUILDKITE_BUILD_NUMBER=$3
 
   # offline inference
   docker exec cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" bash -c "
     set -e
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m"
+    python3 examples/offline_inference/basic.py"
 
   # Run basic model test
   docker exec cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" bash -c "
     set -e
-    pytest -v -s tests/kernels/test_cache.py -m cpu_model
-    pytest -v -s tests/kernels/test_mla_decode_cpu.py -m cpu_model
+    pip install -r vllm/requirements-test.txt
     pytest -v -s tests/models/decoder_only/language -m cpu_model
     pytest -v -s tests/models/embedding/language -m cpu_model
     pytest -v -s tests/models/encoder_decoder/language -m cpu_model
@@ -91,4 +85,4 @@ function cpu_tests() {
 
 # All of CPU tests are expected to be finished less than 40 mins.
 export -f cpu_tests
-timeout 40m bash -c "cpu_tests $CORE_RANGE $NUMA_NODE $BUILDKITE_BUILD_NUMBER"
+timeout 40m bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"

.buildkite/run-gh200-test.sh

@@ -9,13 +9,11 @@ python3 use_existing_torch.py
 
 # Try building the docker image
 DOCKER_BUILDKIT=1 docker build . \
-  --file docker/Dockerfile \
   --target vllm-openai \
   --platform "linux/arm64" \
   -t gh200-test \
   --build-arg max_jobs=66 \
   --build-arg nvcc_threads=2 \
-  --build-arg RUN_WHEEL_CHECK=false \
   --build-arg torch_cuda_arch_list="9.0+PTX" \
   --build-arg vllm_fa_cmake_gpu_arches="90-real"
 
@@ -25,6 +23,6 @@ trap remove_docker_container EXIT
 remove_docker_container
 
 # Run the image and test offline inference
-docker run -e HF_TOKEN -e VLLM_WORKER_MULTIPROC_METHOD=spawn -v /root/.cache/huggingface:/root/.cache/huggingface --name gh200-test --gpus=all --entrypoint="" gh200-test bash -c '
-    python3 examples/offline_inference/basic/generate.py --model meta-llama/Llama-3.2-1B
+docker run -e HF_TOKEN -v /root/.cache/huggingface:/root/.cache/huggingface --name gh200-test --gpus=all --entrypoint="" gh200-test bash -c '
+    python3 examples/offline_inference/cli.py --model meta-llama/Llama-3.2-1B
 '

.buildkite/run-hpu-test.sh

@@ -5,7 +5,7 @@
 set -ex
 
 # Try building the docker image
-docker build -t hpu-test-env -f docker/Dockerfile.hpu .
+docker build -t hpu-test-env -f Dockerfile.hpu .
 
 # Setup cleanup
 # certain versions of HPU software stack have a bug that can
@@ -20,5 +20,5 @@ trap remove_docker_container_and_exit EXIT
 remove_docker_container
 
 # Run the image and launch offline inference
-docker run --runtime=habana --name=hpu-test --network=host -e HABANA_VISIBLE_DEVICES=all -e VLLM_SKIP_WARMUP=true --entrypoint="" hpu-test-env python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m
+docker run --runtime=habana --name=hpu-test --network=host -e HABANA_VISIBLE_DEVICES=all -e VLLM_SKIP_WARMUP=true --entrypoint="" hpu-test-env python3 examples/offline_inference/basic.py
 EXITCODE=$?

.buildkite/run-multi-node-test.sh

@@ -3,7 +3,7 @@
 set -euox pipefail
 
 if [[ $# -lt 4 ]]; then
-    echo "Usage: .buildkite/scripts/run-multi-node-test.sh WORKING_DIR NUM_NODES NUM_GPUS DOCKER_IMAGE COMMAND1 COMMAND2 ... COMMANDN"
+    echo "Usage: .buildkite/run-multi-node-test.sh WORKING_DIR NUM_NODES NUM_GPUS DOCKER_IMAGE COMMAND1 COMMAND2 ... COMMANDN"
     exit 1
 fi
 

.buildkite/run-neuron-test.sh

@@ -35,7 +35,7 @@ else
     date "+%s" > /tmp/neuron-docker-build-timestamp
 fi
 
-docker build -t "${image_name}" -f docker/Dockerfile.neuron .
+docker build -t "${image_name}" -f Dockerfile.neuron .
 
 # Setup cleanup
 remove_docker_container() {
@@ -44,11 +44,11 @@ remove_docker_container() {
 trap remove_docker_container EXIT
 
 # Run the image
-docker run --rm -it --device=/dev/neuron0 --network bridge \
+docker run --rm -it --device=/dev/neuron0 --device=/dev/neuron1 --network host \
        -v "${HF_CACHE}:${HF_MOUNT}" \
        -e "HF_HOME=${HF_MOUNT}" \
        -v "${NEURON_COMPILE_CACHE_URL}:${NEURON_COMPILE_CACHE_MOUNT}" \
        -e "NEURON_COMPILE_CACHE_URL=${NEURON_COMPILE_CACHE_MOUNT}" \
        --name "${container_name}" \
        ${image_name} \
-       /bin/bash -c "python3 /workspace/vllm/examples/offline_inference/neuron.py && python3 -m pytest /workspace/vllm/tests/neuron/1_core/ -v --capture=tee-sys && python3 -m pytest /workspace/vllm/tests/neuron/2_core/ -v --capture=tee-sys"
+       /bin/bash -c "python3 /workspace/vllm/examples/offline_inference/neuron.py && python3 -m pytest /workspace/vllm/tests/neuron/ -v --capture=tee-sys"

.buildkite/run-openvino-test.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+
+# This script build the OpenVINO docker image and run the offline inference inside the container.
+# It serves a sanity check for compilation and basic model usage.
+set -ex
+
+# Try building the docker image
+docker build -t openvino-test -f Dockerfile.openvino .
+
+# Setup cleanup
+remove_docker_container() { docker rm -f openvino-test || true; }
+trap remove_docker_container EXIT
+remove_docker_container
+
+# Run the image and launch offline inference
+docker run --network host --env VLLM_OPENVINO_KVCACHE_SPACE=1 --name openvino-test openvino-test python3 /workspace/examples/offline_inference/basic.py

.buildkite/run-tpu-test.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+
+set -e
+
+# Build the docker image.
+docker build -f Dockerfile.tpu -t vllm-tpu .
+
+# Set up cleanup.
+remove_docker_container() { docker rm -f tpu-test || true; }
+trap remove_docker_container EXIT
+# Remove the container that might not be cleaned up in the previous run.
+remove_docker_container
+
+# For HF_TOKEN.
+source /etc/environment
+# Run a simple end-to-end example.
+docker run --privileged --net host --shm-size=16G -it \
+    -e "HF_TOKEN=$HF_TOKEN" --name tpu-test \
+    vllm-tpu /bin/bash -c "python3 -m pip install git+https://github.com/thuml/depyf.git \
+    && python3 -m pip install pytest \
+    && python3 -m pip install lm_eval[api]==0.4.4 \
+    && pytest -v -s /workspace/vllm/tests/entrypoints/openai/test_accuracy.py \
+    && pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \
+    && python3 /workspace/vllm/tests/tpu/test_compilation.py \
+    && python3 /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
+    && python3 /workspace/vllm/examples/offline_inference/tpu.py"

.buildkite/run-xpu-test.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+# This script build the CPU docker image and run the offline inference inside the container.
+# It serves a sanity check for compilation and basic model usage.
+set -ex
+
+# Try building the docker image
+docker build -t xpu-test -f Dockerfile.xpu .
+
+# Setup cleanup
+remove_docker_container() { docker rm -f xpu-test || true; }
+trap remove_docker_container EXIT
+remove_docker_container
+
+# Run the image and test offline inference/tensor parallel
+docker run --name xpu-test --device /dev/dri -v /dev/dri/by-path:/dev/dri/by-path --entrypoint="" xpu-test sh -c '
+    python3 examples/offline_inference/basic.py
+    python3 examples/offline_inference/cli.py -tp 2
+'

.buildkite/scripts/hardware_ci/run-cpu-test-ppc64le.sh

@@ -1,45 +0,0 @@
-#!/bin/bash
-
-# This script build the CPU docker image and run the offline inference inside the container.
-# It serves a sanity check for compilation and basic model usage.
-set -ex
-
-# Setup cleanup
-remove_docker_container() {
-  if [[ -n "$container_id" ]]; then
-      podman rm -f "$container_id" || true
-  fi
-  podman system prune -f
-}
-trap remove_docker_container EXIT
-remove_docker_container
-
-# Try building the docker image
-podman build -t cpu-test-ubi9-ppc -f docker/Dockerfile.ppc64le .
-
-# Run the image
-container_id=$(podman run -itd --entrypoint /bin/bash -v /tmp/:/root/.cache/huggingface --privileged=true --network host -e HF_TOKEN cpu-test-ubi9-ppc)
-
-function cpu_tests() {
-
-  # offline inference
-  podman exec -it "$container_id" bash -c "
-    set -e
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m"
-
-  # Run basic model test
-  podman exec -it "$container_id" bash -c "
-    set -e
-    pip install pytest pytest-asyncio einops peft Pillow soundfile transformers_stream_generator matplotlib
-    pip install sentence-transformers datamodel_code_generator
-    pytest -v -s tests/models/embedding/language/test_cls_models.py::test_classification_models[float-jason9693/Qwen2.5-1.5B-apeach]
-    pytest -v -s tests/models/embedding/language/test_embedding.py::test_models[half-BAAI/bge-base-en-v1.5]
-    pytest -v -s tests/models/encoder_decoder/language -m cpu_model"
-}
-
-# All of CPU tests are expected to be finished less than 40 mins.
-
-export container_id
-export -f cpu_tests
-timeout 40m bash -c cpu_tests
-

.buildkite/scripts/hardware_ci/run-tpu-v1-test.sh

@@ -1,103 +0,0 @@
-#!/bin/bash
-
-set -xu
-
-# Build the docker image.
-docker build -f docker/Dockerfile.tpu -t vllm-tpu .
-
-# Set up cleanup.
-remove_docker_container() { docker rm -f tpu-test || true; }
-trap remove_docker_container EXIT
-# Remove the container that might not be cleaned up in the previous run.
-remove_docker_container
-
-# For HF_TOKEN.
-source /etc/environment
-# Run a simple end-to-end example.
-docker run --privileged --net host --shm-size=16G -it \
-    -e "HF_TOKEN=$HF_TOKEN" --name tpu-test \
-    vllm-tpu /bin/bash -c "python3 -m pip install git+https://github.com/thuml/depyf.git \
-    && python3 -m pip install pytest pytest-asyncio tpu-info \
-    && python3 -m pip install lm_eval[api]==0.4.4 \
-    && export VLLM_XLA_CACHE_PATH= \
-    && export VLLM_USE_V1=1 \
-    && export VLLM_XLA_CHECK_RECOMPILATION=1 \
-    && echo HARDWARE \
-    && tpu-info \
-    && { \
-        echo TEST_0: Running test_perf.py; \
-        pytest -s -v /workspace/vllm/tests/tpu/test_perf.py; \
-        echo TEST_0_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_1: Running test_compilation.py; \
-        pytest -s -v /workspace/vllm/tests/tpu/test_compilation.py; \
-        echo TEST_1_EXIT_CODE: \$?; \
-    } & \
-    { \
-        echo TEST_2: Running test_basic.py; \
-        pytest -s -v /workspace/vllm/tests/v1/tpu/test_basic.py; \
-        echo TEST_2_EXIT_CODE: \$?; \
-    } & \
-    { \
-        echo TEST_3: Running test_accuracy.py::test_lm_eval_accuracy_v1_engine; \
-        pytest -s -v /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine; \
-        echo TEST_3_EXIT_CODE: \$?; \
-    } & \
-    { \
-        echo TEST_4: Running test_quantization_accuracy.py; \
-        pytest -s -v /workspace/vllm/tests/tpu/test_quantization_accuracy.py; \
-        echo TEST_4_EXIT_CODE: \$?; \
-    } & \
-    { \
-        echo TEST_5: Running examples/offline_inference/tpu.py; \
-        python3 /workspace/vllm/examples/offline_inference/tpu.py; \
-        echo TEST_5_EXIT_CODE: \$?; \
-    } & \
-    { \
-        echo TEST_6: Running test_tpu_model_runner.py; \
-        pytest -s -v /workspace/vllm/tests/tpu/worker/test_tpu_model_runner.py; \
-        echo TEST_6_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_7: Running test_sampler.py; \
-        pytest -s -v /workspace/vllm/tests/v1/tpu/test_sampler.py; \
-        echo TEST_7_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_8: Running test_topk_topp_sampler.py; \
-        pytest -s -v /workspace/vllm/tests/v1/tpu/test_topk_topp_sampler.py; \
-        echo TEST_8_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_9: Running test_multimodal.py; \
-        pytest -s -v /workspace/vllm/tests/v1/tpu/test_multimodal.py; \
-        echo TEST_9_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_10: Running test_pallas.py; \
-        pytest -s -v /workspace/vllm/tests/v1/tpu/test_pallas.py; \
-        echo TEST_10_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_11: Running test_struct_output_generate.py; \
-        pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py; \
-        echo TEST_11_EXIT_CODE: \$?; \
-    } & \
-    && { \
-        echo TEST_12: Running test_moe_pallas.py; \
-        pytest -s -v /workspace/vllm/tests/tpu/test_moe_pallas.py; \
-        echo TEST_12_EXIT_CODE: \$?; \
-    } & \
-    # Disable the TPU LoRA tests until the feature is activated
-    # && { \
-    #     echo TEST_13: Running test_moe_pallas.py; \
-    #     pytest -s -v /workspace/vllm/tests/tpu/lora/; \
-    #     echo TEST_13_EXIT_CODE: \$?; \
-    # } & \
-    wait \
-    && echo 'All tests have attempted to run. Check logs for individual test statuses and exit codes.' \
-"
-
-# TODO: This test fails because it uses RANDOM_SEED sampling
-# && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \

.buildkite/scripts/hardware_ci/run-xpu-test.sh

@@ -1,31 +0,0 @@
-#!/bin/bash
-
-# This script build the CPU docker image and run the offline inference inside the container.
-# It serves a sanity check for compilation and basic model usage.
-set -ex
-
-image_name="xpu/vllm-ci:${BUILDKITE_COMMIT}"
-container_name="xpu_${BUILDKITE_COMMIT}_$(tr -dc A-Za-z0-9 < /dev/urandom | head -c 10; echo)"
-
-# Try building the docker image
-docker build -t ${image_name} -f docker/Dockerfile.xpu .
-
-# Setup cleanup
-remove_docker_container() { 
-  docker rm -f "${container_name}" || true; 
-  docker image rm -f "${image_name}" || true;
-  docker system prune -f || true;
-}
-trap remove_docker_container EXIT
-
-# Run the image and test offline inference/tensor parallel
-docker run \
-    --device /dev/dri \
-    -v /dev/dri/by-path:/dev/dri/by-path \
-    --entrypoint="" \
-    --name "${container_name}" \
-    "${image_name}" \
-    sh -c '
-    VLLM_USE_V1=0 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m
-    VLLM_USE_V1=0 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m -tp 2
-'

.buildkite/test-pipeline.yaml

@@ -2,13 +2,12 @@
 # adding a new command to an existing step. See different options here for examples.
 
 # This script will be feed into Jinja template in `test-template-aws.j2` at
-# https://github.com/vllm-project/buildkite-ci/blob/main/scripts/test-template-aws.j2
+# https://github.com/vllm-project/buildkite-ci/blob/main/scripts/test-template-aws.j2 
 # to generate the final pipeline yaml file.
 
 # Documentation
 # label(str): the name of the test. emoji allowed.
 # fast_check(bool): whether to run this on each commit on fastcheck pipeline.
-# torch_nightly(bool): whether to run this on vllm against torch nightly pipeline.
 # fast_check_only(bool): run this test on fastcheck pipeline only
 # optional(bool): never run this test by default (i.e. need to unblock manually) unless it's scheduled nightly run.
 # command(str): the single command to run for tests. incompatible with commands.
@@ -16,7 +15,7 @@
 # mirror_hardwares(list): the list of hardwares to run the test on as well. currently only supports [amd]
 # gpu(str): override the GPU selection for the test. default is on L4 GPUs. currently only supports a100
 # num_gpus(int): override the number of GPUs for the test. default to 1 GPU. currently support 2,4.
-# num_nodes(int): whether to simulate multi-node setup by launch multiple containers on one host,
+# num_nodes(int): whether to simulate multi-node setup by launch multiple containers on one host, 
 #     in this case, commands must be specified. the first command runs on first host, the second
 #     command runs on the second host.
 # working_dir(str): specify the place where command should execute, default to /vllm-workspace/tests
@@ -25,8 +24,8 @@
 # When adding a test
 # - If the test belong to an existing group, add it there
 # - If the test is short, add to any existing step
-# - If the test takes more than 10min, then it is okay to create a new step.
-#   Note that all steps execute in parallel.
+# - If the test takes more than 10min, then it is okay to create a new step. 
+#   Note that all steps execute in parallel. 
 
 steps:
 ##### fast check tests  #####
@@ -36,12 +35,13 @@ steps:
   fast_check: true
   no_gpu: True
   commands:
-  - pip install -r ../../requirements/docs.txt
+  - pip install -r requirements-docs.txt
   - SPHINXOPTS=\"-W\" make html
   # Check API reference (if it fails, you may have missing mock imports)
-  - grep \"sig sig-object py\" build/html/api/vllm/vllm.sampling_params.html
+  - grep \"sig sig-object py\" build/html/api/inference_params.html
 
 - label: Async Engine, Inputs, Utils, Worker Test # 24min
+  fast_check: true
   source_file_dependencies:
   - vllm/
   - tests/mq_llm_engine
@@ -71,7 +71,6 @@ steps:
 - label: Basic Correctness Test # 30min
   #mirror_hardwares: [amd]
   fast_check: true
-  torch_nightly: true
   source_file_dependencies:
   - vllm/
   - tests/basic_correctness/test_basic_correctness
@@ -79,7 +78,6 @@ steps:
   - tests/basic_correctness/test_preemption
   - tests/basic_correctness/test_cumem.py
   commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
   - pytest -v -s basic_correctness/test_cumem.py
   - pytest -v -s basic_correctness/test_basic_correctness.py
   - pytest -v -s basic_correctness/test_cpu_offload.py
@@ -106,8 +104,7 @@ steps:
 - label: Entrypoints Test # 40min
   working_dir: "/vllm-workspace/tests"
   fast_check: true
-  torch_nightly: true
-  #mirror_hardwares: [amd]
+  mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/
   - tests/entrypoints/llm
@@ -115,19 +112,19 @@ steps:
   - tests/entrypoints/test_chat_utils
   - tests/entrypoints/offline_mode
   commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
   - pytest -v -s entrypoints/llm --ignore=entrypoints/llm/test_lazy_outlines.py --ignore=entrypoints/llm/test_generate.py --ignore=entrypoints/llm/test_generate_multiple_loras.py --ignore=entrypoints/llm/test_guided_generate.py --ignore=entrypoints/llm/test_collective_rpc.py
   - pytest -v -s entrypoints/llm/test_lazy_outlines.py # it needs a clean process
   - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process
   - pytest -v -s entrypoints/llm/test_generate_multiple_loras.py # it needs a clean process
-  - VLLM_USE_V1=0 pytest -v -s entrypoints/llm/test_guided_generate.py # it needs a clean process
-  - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_oot_registration.py  --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/test_openai_schema.py
+  - pytest -v -s entrypoints/llm/test_guided_generate.py # it needs a clean process
+  - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_oot_registration.py
   - pytest -v -s entrypoints/test_chat_utils.py
-  - VLLM_USE_V1=0 pytest -v -s entrypoints/offline_mode # Needs to avoid interference with other tests
+  - pytest -v -s entrypoints/offline_mode # Needs to avoid interference with other tests
 
 - label: Distributed Tests (4 GPUs) # 10min
   working_dir: "/vllm-workspace/tests"
   num_gpus: 4
+  fast_check: true
   source_file_dependencies:
   - vllm/distributed/
   - vllm/core/
@@ -137,42 +134,37 @@ steps:
   - tests/compile/test_basic_correctness
   - examples/offline_inference/rlhf.py
   - examples/offline_inference/rlhf_colocate.py
-  - tests/examples/offline_inference/data_parallel.py
-  - tests/v1/test_async_llm_dp.py
   commands:
-  # test with tp=2 and external_dp=2
-  - VLLM_USE_V1=0 torchrun --nproc-per-node=4 distributed/test_torchrun_example.py
-  - torchrun --nproc-per-node=4 distributed/test_torchrun_example.py
-  # test with internal dp
-  - python3 ../examples/offline_inference/data_parallel.py
-  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/test_async_llm_dp.py
   - pytest -v -s distributed/test_utils.py
   - pytest -v -s compile/test_basic_correctness.py
   - pytest -v -s distributed/test_pynccl.py
   - pytest -v -s spec_decode/e2e/test_integration_dist_tp4.py
   # TODO: create a dedicated test section for multi-GPU example tests
   # when we have multiple distributed example tests
-  - pushd ../examples/offline_inference
-  - python3 rlhf.py
-  - RAY_DEDUP_LOGS=0 python3 rlhf_colocate.py
-  - popd
+  - python3 ../examples/offline_inference/rlhf.py
+  - RAY_DEDUP_LOGS=0 python3 ../examples/offline_inference/rlhf_colocate.py
 
 - label: Metrics, Tracing Test # 10min
-  mirror_hardwares: [amd]
-  num_gpus: 2
+  num_gpus: 2 
+  fast_check: true
   source_file_dependencies:
   - vllm/
   - tests/metrics
   - tests/tracing
   commands:
-  - pytest -v -s metrics
+  - pytest -v -s metrics 
+  - "pip install \
+      'opentelemetry-sdk>=1.26.0,<1.27.0' \
+      'opentelemetry-api>=1.26.0,<1.27.0' \
+      'opentelemetry-exporter-otlp>=1.26.0,<1.27.0' \
+      'opentelemetry-semantic-conventions-ai>=0.4.1,<0.5.0'"
   - pytest -v -s tracing
 
 ##### fast check tests  #####
 #####  1 GPU test  #####
 
 - label: Regression Test # 5min
-  #mirror_hardwares: [amd]
+  mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/
   - tests/test_regression
@@ -202,23 +194,18 @@ steps:
     - tests/v1
   commands:
     # split the test to avoid interference
-    - pytest -v -s v1/core
-    - pytest -v -s v1/engine
-    - pytest -v -s v1/entrypoints
-    - pytest -v -s v1/sample
-    - pytest -v -s v1/worker
-    - pytest -v -s v1/structured_output
-    - pytest -v -s v1/spec_decode
-    - pytest -v -s v1/test_serial_utils.py
-    - pytest -v -s v1/test_stats.py
-    - pytest -v -s v1/test_utils.py
-    - pytest -v -s v1/test_oracle.py
+    - VLLM_USE_V1=1 pytest -v -s v1/core
+    - VLLM_USE_V1=1 pytest -v -s v1/engine
+    - VLLM_USE_V1=1 pytest -v -s v1/sample
+    - VLLM_USE_V1=1 pytest -v -s v1/worker
+    - VLLM_USE_V1=1 pytest -v -s v1/test_stats.py
+    - VLLM_USE_V1=1 pytest -v -s v1/test_utils.py
     # TODO: accuracy does not match, whether setting
     # VLLM_USE_FLASHINFER_SAMPLER or not on H100.
-    - pytest -v -s v1/e2e
+    - VLLM_USE_V1=1 pytest -v -s v1/e2e
     # Integration test for streaming correctness (requires special branch).
     - pip install -U git+https://github.com/robertgshaw2-neuralmagic/lm-evaluation-harness.git@streaming-api
-    - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
+    - pytest -v -s entrypoints/openai/test_accuracy.py::test_lm_eval_accuracy_v1_engine
 
 - label: Examples Test # 25min
   working_dir: "/vllm-workspace/examples"
@@ -228,22 +215,19 @@ steps:
   - examples/
   commands:
     - pip install tensorizer # for tensorizer test
-    - python3 offline_inference/basic/generate.py --model facebook/opt-125m
-    - python3 offline_inference/basic/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
-    - python3 offline_inference/basic/chat.py
+    - python3 offline_inference/basic.py
+    - python3 offline_inference/cpu_offload.py
+    - python3 offline_inference/chat.py
     - python3 offline_inference/prefix_caching.py
     - python3 offline_inference/llm_engine_example.py
-    - python3 offline_inference/audio_language.py --seed 0
-    - python3 offline_inference/vision_language.py --seed 0
-    - python3 offline_inference/vision_language_embedding.py --seed 0
-    - python3 offline_inference/vision_language_multi_image.py --seed 0
-    - VLLM_USE_V1=0 python3 other/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 other/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
+    - python3 offline_inference/vision_language.py
+    - python3 offline_inference/vision_language_multi_image.py
+    - python3 other/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 other/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
     - python3 offline_inference/encoder_decoder.py
-    - python3 offline_inference/encoder_decoder_multimodal.py --model-type whisper --seed 0
-    - python3 offline_inference/basic/classify.py
-    - python3 offline_inference/basic/embed.py
-    - python3 offline_inference/basic/score.py
-    - VLLM_USE_V1=0 python3 offline_inference/profiling.py --model facebook/opt-125m run_num_steps --num-steps 2
+    - python3 offline_inference/classification.py
+    - python3 offline_inference/embedding.py
+    - python3 offline_inference/scoring.py
+    - python3 offline_inference/profiling.py --model facebook/opt-125m run_num_steps --num-steps 2
 
 - label: Prefix Caching Test # 9min
   mirror_hardwares: [amd]
@@ -270,7 +254,7 @@ steps:
   - vllm/model_executor/guided_decoding
   - tests/test_logits_processor
   - tests/model_executor/test_guided_processors
-  commands:
+  commands: 
     - pytest -v -s test_logits_processor.py
     - pytest -v -s model_executor/test_guided_processors.py
 
@@ -281,29 +265,19 @@ steps:
   - vllm/model_executor/models/eagle.py
   commands:
     - pytest -v -s spec_decode/e2e/test_multistep_correctness.py
-    - VLLM_ATTENTION_BACKEND=FLASH_ATTN pytest -v -s spec_decode --ignore=spec_decode/e2e/test_multistep_correctness.py --ignore=spec_decode/e2e/test_mtp_correctness.py
+    - VLLM_ATTENTION_BACKEND=FLASH_ATTN pytest -v -s spec_decode --ignore=spec_decode/e2e/test_multistep_correctness.py
     - pytest -v -s spec_decode/e2e/test_eagle_correctness.py
 
 - label: LoRA Test %N # 15min each
-  #mirror_hardwares: [amd]
+  mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/lora
   - tests/lora
-  command: pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_chatglm3_tp.py --ignore=lora/test_llama_tp.py
+  command: pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_long_context.py --ignore=lora/test_chatglm3_tp.py --ignore=lora/test_llama_tp.py --ignore=lora/test_minicpmv_tp.py
   parallelism: 4
 
-- label: PyTorch Compilation Unit Tests
-  torch_nightly: true
-  source_file_dependencies:
-    - vllm/
-    - tests/compile
-  commands:
-    - pytest -v -s compile/test_pass_manager.py
-    - pytest -v -s compile/test_fusion.py
-    - pytest -v -s compile/test_sequence_parallelism.py
-
-- label: PyTorch Fullgraph Smoke Test # 9min
-  torch_nightly: true
+- label: "PyTorch Fullgraph Smoke Test" # 9min
+  fast_check: true
   source_file_dependencies:
   - vllm/
   - tests/compile
@@ -313,62 +287,25 @@ steps:
   - pytest -v -s compile/piecewise/test_simple.py
   - pytest -v -s compile/piecewise/test_toy_llama.py
 
-- label: PyTorch Fullgraph Test # 18min
-  torch_nightly: true
+- label: "PyTorch Fullgraph Test" # 18min
   source_file_dependencies:
   - vllm/
   - tests/compile
   commands:
   - pytest -v -s compile/test_full_graph.py
 
-- label: Kernels Core Operation Test
+- label: Kernels Test %N # 1h each
   mirror_hardwares: [amd]
   source_file_dependencies:
   - csrc/
-  - tests/kernels/core
-  commands:
-    - pytest -v -s kernels/core
-
-- label: Kernels Attention Test %N
-  mirror_hardwares: [amd]
-  source_file_dependencies:
-  - csrc/attention/
   - vllm/attention
-  - vllm/v1/attention
-  - tests/kernels/attention
+  - tests/kernels
   commands:
-    - pytest -v -s kernels/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
-  parallelism: 2
-
-- label: Kernels Quantization Test %N
-  mirror_hardwares: [amd]
-  source_file_dependencies:
-  - csrc/quantization/
-  - vllm/model_executor/layers/quantization
-  - tests/kernels/quantization
-  commands:
-    - pytest -v -s kernels/quantization  --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
-  parallelism: 2
-
-- label: Kernels MoE Test
-  #mirror_hardwares: [amd]
-  source_file_dependencies:
-  - csrc/moe/
-  - tests/kernels/moe
-  - vllm/model_executor/layers/fused_moe/
-  commands:
-    - pytest -v -s kernels/moe
-
-- label: Kernels Mamba Test
-  #mirror_hardwares: [amd]
-  source_file_dependencies:
-  - csrc/mamba/
-  - tests/kernels/mamba
-  commands:
-    - pytest -v -s kernels/mamba
+    - pytest -v -s kernels --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
+  parallelism: 4
 
 - label: Tensorizer Test # 11min
-  # mirror_hardwares: [amd]
+  mirror_hardwares: [amd]
   soft_fail: true
   source_file_dependencies:
   - vllm/model_executor/model_loader
@@ -384,22 +321,14 @@ steps:
   source_file_dependencies:
   - benchmarks/
   commands:
-  - bash scripts/run-benchmarks.sh
+  - bash run-benchmarks.sh
 
-- label: Benchmarks CLI Test # 10min
-  source_file_dependencies:
-  - vllm/
-  - tests/benchmarks/
-  commands:
-  - pytest -v -s benchmarks/
-
-- label: Quantization Test
+- label: Quantization Test # 33min
   source_file_dependencies:
   - csrc/
   - vllm/model_executor/layers/quantization
   - tests/quantization
-  commands:
-  - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization
+  command: VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization
 
 - label: LM Eval Small Models # 53min
   working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
@@ -408,15 +337,7 @@ steps:
   - vllm/model_executor/layers/quantization
   commands:
   - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-small.txt --tp-size=1
-
-- label: OpenAI API correctness
-  source_file_dependencies:
-  - csrc/
-  - vllm/entrypoints/openai/
-  - vllm/model_executor/models/whisper.py
-  commands: # LMEval+Transcription WER check
-  - pytest -s entrypoints/openai/correctness/
+  - bash ./run-tests.sh -c configs/models-small.txt -t 1
 
 - label: Encoder Decoder tests # 5min
   source_file_dependencies:
@@ -427,101 +348,96 @@ steps:
 
 - label: OpenAI-Compatible Tool Use # 20 min
   fast_check: false
-  #mirror_hardwares: [ amd ]
+  mirror_hardwares: [ amd ]
   source_file_dependencies:
     - vllm/
     - tests/tool_use
-    - tests/mistral_tool_use
   commands:
     - pytest -v -s tool_use
-    - pytest -v -s mistral_tool_use
 
 #####  models test  #####
 
 - label: Basic Models Test # 24min
-  torch_nightly: true
   source_file_dependencies:
   - vllm/
   - tests/models
   commands:
     - pytest -v -s models/test_transformers.py
     - pytest -v -s models/test_registry.py
-    - pytest -v -s models/test_utils.py
-    - pytest -v -s models/test_vision.py
-    # V1 Test: https://github.com/vllm-project/vllm/issues/14531
-    - VLLM_USE_V1=0 pytest -v -s models/test_initialization.py -k 'not llama4 and not plamo2'
-    - VLLM_USE_V1=0 pytest -v -s models/test_initialization.py -k 'llama4'
-    - VLLM_USE_V1=0 pytest -v -s models/test_initialization.py -k 'plamo2'
+    - pytest -v -s models/test_initialization.py
 
-- label: Language Models Test (Standard)
+- label: Language Models Test (Standard) # 32min
   #mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/
-  - tests/models/language
+  - tests/models/decoder_only/language
+  - tests/models/embedding/language
+  - tests/models/encoder_decoder/language
   commands:
-    # Install causal-conv1d for plamo2 models here, as it is not compatible with pip-compile.
-    - pip install 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.0.post8'
-    - pytest -v -s models/language -m core_model
+    - pytest -v -s models/decoder_only/language -m 'core_model or quant_model'
+    - pytest -v -s models/embedding/language -m core_model
 
-- label: Language Models Test (Extended)
+- label: Language Models Test (Extended) # 1h10min
   optional: true
   source_file_dependencies:
   - vllm/
-  - tests/models/language
+  - tests/models/decoder_only/language
+  - tests/models/embedding/language
+  - tests/models/encoder_decoder/language
   commands:
-    # Install causal-conv1d for plamo2 models here, as it is not compatible with pip-compile.
-    - pip install 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.0.post8'
-    - pytest -v -s models/language -m 'not core_model'
+    - pytest -v -s models/decoder_only/language -m 'not core_model and not quant_model'
+    - pytest -v -s models/embedding/language -m 'not core_model'
 
-- label: Multi-Modal Models Test (Standard)
+- label: Multi-Modal Models Test (Standard) # 40min
   #mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/
-  - tests/models/multimodal
+  - tests/models/decoder_only/audio_language
+  - tests/models/decoder_only/vision_language
+  - tests/models/embedding/vision_language
+  - tests/models/encoder_decoder/audio_language
+  - tests/models/encoder_decoder/vision_language
   commands:
     - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/processing
-    - pytest -v -s --ignore models/multimodal/generation/test_whisper.py models/multimodal -m core_model
-    - cd .. && pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model  # Otherwise, mp_method="spawn" doesn't work
+    - pytest -v -s models/multimodal
+    - pytest -v -s models/decoder_only/audio_language -m 'core_model or quant_model'
+    - pytest -v -s --ignore models/decoder_only/vision_language/test_phi3v.py models/decoder_only/vision_language -m 'core_model or quant_model'
+    - pytest -v -s models/embedding/vision_language -m core_model
+    - pytest -v -s models/encoder_decoder/audio_language -m core_model
+    - pytest -v -s models/encoder_decoder/language -m core_model
+    - pytest -v -s models/encoder_decoder/vision_language -m core_model
 
-- label: Multi-Modal Models Test (Extended) 1
+- label: Multi-Modal Models Test (Extended) 1 # 48m
   optional: true
   source_file_dependencies:
   - vllm/
-  - tests/models/multimodal
+  - tests/models/decoder_only/audio_language
+  - tests/models/decoder_only/vision_language
+  - tests/models/embedding/vision_language
+  - tests/models/encoder_decoder/vision_language
   commands:
     - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/processing models/multimodal -m 'not core_model'
+    - pytest -v -s models/decoder_only/audio_language -m 'not core_model and not quant_model'
+    - pytest -v -s models/decoder_only/vision_language/test_models.py -m 'split(group=0) and not core_model and not quant_model'
+    # HACK - run phi3v tests separately to sidestep this transformers bug
+    # https://github.com/huggingface/transformers/issues/34307
+    - pytest -v -s models/decoder_only/vision_language/test_phi3v.py
+    - pytest -v -s --ignore models/decoder_only/vision_language/test_models.py --ignore models/decoder_only/vision_language/test_phi3v.py models/decoder_only/vision_language -m 'not core_model and not quant_model'
+    - pytest -v -s models/embedding/vision_language -m 'not core_model'
+    - pytest -v -s models/encoder_decoder/language -m 'not core_model'
+    - pytest -v -s models/encoder_decoder/vision_language -m 'not core_model'
 
-- label: Multi-Modal Models Test (Extended) 2
+- label: Multi-Modal Models Test (Extended) 2 # 38m
   optional: true
   source_file_dependencies:
   - vllm/
-  - tests/models/multimodal
+  - tests/models/decoder_only/vision_language
   commands:
     - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=0) and not core_model'
-
-- label: Multi-Modal Models Test (Extended) 3
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
-
-- label: Quantized Models Test
-  #mirror_hardwares: [amd]
-  source_file_dependencies:
-  - vllm/model_executor/layers/quantization
-  - tests/models/quantization
-  commands:
-    - pytest -v -s models/quantization
+    - pytest -v -s models/decoder_only/vision_language/test_models.py -m 'split(group=1) and not core_model and not quant_model'
 
 # This test is used only in PR development phase to test individual models and should never run on main
 - label: Custom Models Test
-  mirror_hardwares: [amd]
   optional: true
   commands:
     - echo 'Testing custom models...'
@@ -533,7 +449,6 @@ steps:
 #####  multi gpus test  #####
 
 - label: Distributed Comm Ops Test # 7min
-  mirror_hardwares: [amd]
   working_dir: "/vllm-workspace/tests"
   num_gpus: 2
   source_file_dependencies:
@@ -576,31 +491,28 @@ steps:
   - vllm/worker/worker.py
   - vllm/worker/model_runner.py
   - entrypoints/llm/test_collective_rpc.py
-  - tests/v1/test_async_llm_dp.py
-  - vllm/v1/engine/
   commands:
-  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/test_async_llm_dp.py
   - pytest -v -s entrypoints/llm/test_collective_rpc.py
+  - torchrun --nproc-per-node=2 distributed/test_torchrun_example.py
   - pytest -v -s ./compile/test_basic_correctness.py
   - pytest -v -s ./compile/test_wrapper.py
   - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
   - TARGET_TEST_SUITE=L4 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
   # Avoid importing model tests that cause CUDA reinitialization error
   - pytest models/test_transformers.py -v -s -m 'distributed(num_gpus=2)'
-  - pytest models/language -v -s -m 'distributed(num_gpus=2)'
-  - pytest models/multimodal -v -s -m 'distributed(num_gpus=2)'
-  # test sequence parallel
-  - pytest -v -s distributed/test_sequence_parallel.py
+  - pytest models/encoder_decoder/language/test_bart.py -v -s -m 'distributed(num_gpus=2)'
+  - pytest models/encoder_decoder/vision_language/test_broadcast.py -v -s -m 'distributed(num_gpus=2)'
+  - pytest models/decoder_only/vision_language/test_models.py -v -s -m 'distributed(num_gpus=2)'
   # this test fails consistently.
   # TODO: investigate and fix
   # - pytest -v -s spec_decode/e2e/test_integration_dist_tp2.py
-  - VLLM_USE_V1=0 CUDA_VISIBLE_DEVICES=0,1 pytest -v -s test_sharded_state_loader.py
-  - VLLM_USE_V1=0 CUDA_VISIBLE_DEVICES=0,1 pytest -v -s kv_transfer/test_disagg.py
-  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown
+  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s test_sharded_state_loader.py
+  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s kv_transfer/disagg_test.py
 
 - label: Plugin Tests (2 GPUs) # 40min
   working_dir: "/vllm-workspace/tests"
   num_gpus: 2
+  fast_check: true
   source_file_dependencies:
   - vllm/plugins/
   - tests/plugins/
@@ -611,7 +523,6 @@ steps:
   - pip uninstall vllm_add_dummy_platform -y
   # end platform plugin tests
   # other tests continue here:
-  - pytest -v -s plugins_tests/test_scheduler_plugins.py
   - pip install -e ./plugins/vllm_add_dummy_model
   - pytest -v -s distributed/test_distributed_oot.py
   - pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process
@@ -659,10 +570,13 @@ steps:
     # FIXIT: find out which code initialize cuda before running the test
     # before the fix, we need to use spawn to test it
     - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    # There is some Tensor Parallelism related processing logic in LoRA that
+    # This test runs llama 13B, so it is required to run on 4 GPUs.
+    - pytest -v -s -x lora/test_long_context.py
+    # There is some Tensor Parallelism related processing logic in LoRA that 
     # requires multi-GPU testing for validation.
     - pytest -v -s -x lora/test_chatglm3_tp.py
     - pytest -v -s -x lora/test_llama_tp.py
+    - pytest -v -s -x lora/test_minicpmv_tp.py
 
 
 - label: Weight Loading Multiple GPU Test  # 33min
@@ -683,7 +597,7 @@ steps:
   - vllm/
   - tests/weight_loading
   commands:
-    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large.txt
+    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large.txt 
 
 
 ##### multi gpus test #####
@@ -695,7 +609,7 @@ steps:
   num_gpus: 4
   source_file_dependencies:
   - vllm/
-  commands:
+  commands: 
   # NOTE: don't test llama model here, it seems hf implementation is buggy
   # see https://github.com/vllm-project/vllm/pull/5689 for details
   - pytest -v -s distributed/test_custom_all_reduce.py
@@ -713,4 +627,4 @@ steps:
   - vllm/model_executor/layers/quantization
   commands:
   - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
+  - bash ./run-tests.sh -c configs/models-large.txt -t 4

.buildkite/upload-wheels.sh

@@ -50,11 +50,8 @@ aws s3 cp "$normal_wheel" "s3://vllm-wheels/$BUILDKITE_COMMIT/"
 if [[ $normal_wheel == *"cu118"* ]]; then
     # if $normal_wheel matches cu118, do not upload the index.html
     echo "Skipping index files for cu118 wheels"
-elif [[ $normal_wheel == *"cu126"* ]]; then
-    # if $normal_wheel matches cu126, do not upload the index.html
-    echo "Skipping index files for cu126 wheels"
 else
-    # only upload index.html for cu128 wheels (default wheels)
+    # only upload index.html for cu12 wheels (default wheels)
     aws s3 cp index.html "s3://vllm-wheels/$BUILDKITE_COMMIT/vllm/index.html"
     aws s3 cp "s3://vllm-wheels/nightly/index.html" "s3://vllm-wheels/$BUILDKITE_COMMIT/index.html"
 fi
@@ -66,12 +63,9 @@ aws s3 cp "$normal_wheel" "s3://vllm-wheels/nightly/"
 if [[ $normal_wheel == *"cu118"* ]]; then
     # if $normal_wheel matches cu118, do not upload the index.html
     echo "Skipping index files for cu118 wheels"
-elif [[ $normal_wheel == *"cu126"* ]]; then
-    # if $normal_wheel matches cu126, do not upload the index.html
-    echo "Skipping index files for cu126 wheels"
 else
-    # only upload index.html for cu128 wheels (default wheels)
+    # only upload index.html for cu12 wheels (default wheels)
     aws s3 cp index.html "s3://vllm-wheels/nightly/vllm/index.html"
 fi
 
-aws s3 cp "$wheel" "s3://vllm-wheels/$version/"
+aws s3 cp "$wheel" "s3://vllm-wheels/$version/"

.github/CODEOWNERS

@@ -10,33 +10,27 @@
 /vllm/worker/worker.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
 /vllm/model_executor/layers/sampler.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
 /vllm/model_executor/layers/quantization @mgoin @robertgshaw2-redhat @tlrmchlsmth
-/vllm/model_executor/guided_decoding @mgoin @russellb
+/vllm/model_executor/guided_decoding @mgoin
 /vllm/multimodal @DarkLight1337 @ywang96
-/vllm/vllm_flash_attn @LucasWilkinson
 CMakeLists.txt @tlrmchlsmth
 
 # vLLM V1
 /vllm/v1 @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @comaniac @alexm-redhat
-/vllm/v1/structured_output @mgoin @russellb
 
 # Test ownership
-/.buildkite/lm-eval-harness @mgoin @simon-mo
 /tests/async_engine @njhill @robertgshaw2-redhat @simon-mo
-/tests/basic_correctness/test_chunked_prefill @rkooo567 @comaniac
+/tests/test_inputs.py @DarkLight1337 @ywang96
+/tests/entrypoints @DarkLight1337 @robertgshaw2-redhat @simon-mo
+/tests/models @DarkLight1337 @ywang96
+/tests/multimodal @DarkLight1337 @ywang96
+/tests/prefix_caching @comaniac @KuntaiDu
+/tests/spec_decode @njhill @LiuXiaoxuanPKU
+/tests/kernels @tlrmchlsmth @WoosukKwon
+/tests/quantization @mgoin @robertgshaw2-redhat
+/.buildkite/lm-eval-harness @mgoin @simon-mo
 /tests/distributed/test_multi_node_assignment.py @youkaichao
 /tests/distributed/test_pipeline_parallel.py @youkaichao
 /tests/distributed/test_same_node.py @youkaichao
-/tests/entrypoints @DarkLight1337 @robertgshaw2-redhat @simon-mo
-/tests/entrypoints/llm/test_guided_generate.py @mgoin @russellb
-/tests/kernels @tlrmchlsmth @WoosukKwon
-/tests/model_executor/test_guided_processors.py @mgoin @russellb
-/tests/models @DarkLight1337 @ywang96
 /tests/multi_step @alexm-redhat @comaniac
-/tests/multimodal @DarkLight1337 @ywang96
-/tests/prefix_caching @comaniac @KuntaiDu
-/tests/quantization @mgoin @robertgshaw2-redhat
-/tests/spec_decode @njhill @LiuXiaoxuanPKU
-/tests/test_inputs.py @DarkLight1337 @ywang96
-/tests/v1/entrypoints/llm/test_struct_output_generate.py @mgoin @russellb
-/tests/v1/structured_output @mgoin @russellb
 /tests/weight_loading @mgoin @youkaichao
+/tests/basic_correctness/test_chunked_prefill @rkooo567 @comaniac

.github/ISSUE_TEMPLATE/200-installation.yml

@@ -14,7 +14,7 @@ body:
     description: |
       Please run the following and paste the output below.
       ```sh
-      wget https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/collect_env.py
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
       # For security purposes, please feel free to check the contents of collect_env.py before running it.
       python collect_env.py
       ```

.github/ISSUE_TEMPLATE/300-usage.yml

@@ -14,7 +14,7 @@ body:
     description: |
       Please run the following and paste the output below.
       ```sh
-      wget https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/collect_env.py
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
       # For security purposes, please feel free to check the contents of collect_env.py before running it.
       python collect_env.py
       ```

.github/ISSUE_TEMPLATE/400-bug-report.yml

@@ -14,19 +14,19 @@ body:
     description: |
       Please run the following and paste the output below.
       ```sh
-      wget https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/collect_env.py
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
       # For security purposes, please feel free to check the contents of collect_env.py before running it.
       python collect_env.py
       ```
       It is suggested to download and execute the latest script, as vllm might frequently update the diagnosis information needed for accurately and quickly responding to issues.
     value: |
       <details>
-      <summary>The output of <code>python collect_env.py</code></summary>
+      <summary>The output of `python collect_env.py`</summary>
 
       ```text
       Your output of `python collect_env.py` here
       ```
-
+      
       </details>
   validations:
     required: true
@@ -75,7 +75,7 @@ body:
       ```
 
       ```
-      The error message you got, with the full traceback and the error logs with [dump_input.py:##] if present.
+      The error message you got, with the full traceback.
       ```
   validations:
     required: true

.github/ISSUE_TEMPLATE/600-new-model.yml

@@ -9,7 +9,7 @@ body:
     value: >
       #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
 
-      #### We also highly recommend you read https://docs.vllm.ai/en/latest/contributing/model/index.html first to understand how to add a new model.
+      #### We also highly recommend you read https://docs.vllm.ai/en/latest/contributing/model/adding_model.html first to understand how to add a new model.
 - type: textarea
   attributes:
     label: The model to consider.

.github/ISSUE_TEMPLATE/700-performance-discussion.yml

@@ -35,7 +35,7 @@ body:
     description: |
       Please run the following and paste the output below.
       ```sh
-      wget https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/collect_env.py
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
       # For security purposes, please feel free to check the contents of collect_env.py before running it.
       python collect_env.py
       ```

.github/ISSUE_TEMPLATE/800-misc-discussion.yml

@@ -0,0 +1,28 @@
+name: 🎲 Misc/random discussions that do not fit into the above categories.
+description: Submit a discussion as you like. Note that developers are heavily overloaded and we mainly rely on community users to answer these issues.
+title: "[Misc]: "
+labels: ["misc"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Anything you want to discuss about vllm.
+    description: >
+      Anything you want to discuss about vllm.
+  validations:
+    required: true
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!
+- type: checkboxes
+  id: askllm
+  attributes:
+    label: Before submitting a new issue...
+    options:
+      - label: Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://docs.vllm.ai/en/latest/), which can answer lots of frequently asked questions.
+        required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -1,5 +1 @@
 blank_issues_enabled: false
-contact_links:
-  - name: Questions
-    url: https://discuss.vllm.ai
-    about: Ask questions and discuss with other vLLM community members

.github/PULL_REQUEST_TEMPLATE.md

@@ -3,4 +3,4 @@ FILL IN THE PR DESCRIPTION HERE
 FIX #xxxx (*link existing issues this PR will resolve*)
 
 <!--- pyml disable-next-line no-emphasis-as-heading -->
-**BEFORE SUBMITTING, PLEASE READ <https://docs.vllm.ai/en/latest/contributing/overview.html>** (anything written below this line will be removed by GitHub Actions)
+**BEFORE SUBMITTING, PLEASE READ <https://docs.vllm.ai/en/latest/contributing/overview.html>**

.github/dependabot.yml

@@ -23,7 +23,7 @@ updates:
       - dependency-name: "lm-format-enforcer"
       - dependency-name: "gguf"
       - dependency-name: "compressed-tensors"
-      - dependency-name: "ray[cgraph]" # Ray Compiled Graph
+      - dependency-name: "ray[adag]"
       - dependency-name: "lm-eval"
     groups:
       minor-update:

.github/mergify.yml

@@ -5,7 +5,6 @@ pull_request_rules:
     - or:
       - files~=^[^/]+\.md$
       - files~=^docs/
-      - files~=^examples/
   actions:
     label:
       add:
@@ -19,7 +18,7 @@ pull_request_rules:
       - files~=\.buildkite/
       - files~=^cmake/
       - files=CMakeLists.txt
-      - files~=^docker/Dockerfile
+      - files~=^Dockerfile
       - files~=^requirements.*\.txt
       - files=setup.py
   actions:
@@ -36,38 +35,15 @@ pull_request_rules:
       add:
         - frontend
 
-- name: label-multi-modality
-  description: Automatically apply multi-modality label
-  conditions:
-    - or:
-      - files~=^vllm/multimodal/
-      - files~=^tests/multimodal/
-      - files~=^tests/models/multimodal/
-      - files~=^tests/models/*/audio_language/
-      - files~=^tests/models/*/vision_language/
-      - files=tests/models/test_vision.py
-  actions:
-    label:
-      add:
-        - multi-modality
-
 - name: label-structured-output
   description: Automatically apply structured-output label
   conditions:
     - or:
-      - files~=^benchmarks/structured_schemas/
-      - files=benchmarks/benchmark_serving_structured_output.py
-      - files=benchmarks/run_structured_output_benchmark.sh
-      - files=docs/source/features/structured_outputs.md
-      - files=examples/offline_inference/structured_outputs.py
-      - files=examples/online_serving/openai_chat_completion_structured_outputs.py
-      - files=examples/online_serving/openai_chat_completion_structured_outputs_with_reasoning.py
       - files~=^vllm/model_executor/guided_decoding/
       - files=tests/model_executor/test_guided_processors.py
       - files=tests/entrypoints/llm/test_guided_generate.py
-      - files~=^tests/v1/structured_output/
-      - files=tests/v1/entrypoints/llm/test_guided_generate.py
-      - files~=^vllm/v1/structured_output/
+      - files=benchmarks/benchmark_serving_guided.py
+      - files=benchmarks/benchmark_guided.py
   actions:
     label:
       add:
@@ -96,58 +72,6 @@ pull_request_rules:
       add:
         - v1
 
-- name: label-tpu
-  description: Automatically apply tpu label
-  # Keep this list in sync with `label-tpu-remove` conditions
-  conditions:
-    - or:
-      - files~=tpu.py
-      - files~=_tpu
-      - files~=tpu_
-      - files~=/tpu/
-      - files~=pallas
-  actions:
-    label:
-      add:
-        - tpu
-
-- name: label-tpu-remove
-  description: Automatically remove tpu label
-  # Keep this list in sync with `label-tpu` conditions
-  conditions:
-    - and:
-      - -files~=tpu.py
-      - -files~=_tpu
-      - -files~=tpu_
-      - -files~=/tpu/
-      - -files~=pallas
-  actions:
-    label:
-      remove:
-        - tpu
-
-- name: label-tool-calling
-  description: Automatically add tool-calling label
-  conditions:
-    - or:
-      - files~=^tests/tool_use/
-      - files~=^tests/mistral_tool_use/
-      - files~=^tests/entrypoints/openai/tool_parsers/
-      - files=tests/entrypoints/openai/test_chat_with_tool_reasoning.py
-      - files~=^vllm/entrypoints/openai/tool_parsers/
-      - files=docs/source/features/tool_calling.md
-      - files=docs/source/getting_started/examples/openai_chat_completion_client_with_tools.md
-      - files=docs/source/getting_started/examples/chat_with_tools.md
-      - files~=^examples/tool_chat_*
-      - files=examples/offline_inference/chat_with_tools.py
-      - files=examples/online_serving/openai_chat_completion_client_with_tools_required.py
-      - files=examples/online_serving/openai_chat_completion_tool_calls_with_reasoning.py
-      - files=examples/online_serving/openai_chat_completion_client_with_tools.py
-  actions:
-    label:
-      add:
-        - tool-calling
-
 - name: ping author on conflicts and add 'needs-rebase' label
   conditions:
       - conflict

.github/workflows/lint-and-deploy.yaml

@@ -12,7 +12,7 @@ jobs:
           fetch-depth: 0
 
       - name: Set up Helm
-        uses: azure/setup-helm@b9e51907a09c216f16ebe8536097933489208112 # v4.3.0
+        uses: azure/setup-helm@fe7b79cd5ee1e45176fcad797de68ecaf3ca4814 # v4.2.0
         with:
           version: v3.14.4
 
@@ -50,7 +50,7 @@ jobs:
         uses: helm/kind-action@a1b0e391336a6ee6713a0583f8c6240d70863de3 # v1.12.0
 
       - name: Build the Docker image vllm cpu
-        run: docker buildx build -f docker/Dockerfile.cpu -t vllm-cpu-env .
+        run: docker buildx build -f Dockerfile.cpu -t vllm-cpu-env .
 
       - name: Configuration of docker images, network and namespace for the kind cluster
         run: |
@@ -66,7 +66,7 @@ jobs:
           export AWS_SECRET_ACCESS_KEY=minioadmin
           sleep 30 && kubectl -n ns-vllm logs -f "$(kubectl -n ns-vllm get pods | awk '/deployment/ {print $1;exit}')" &
           helm install --wait --wait-for-jobs --timeout 5m0s --debug --create-namespace --namespace=ns-vllm test-vllm examples/online_serving/chart-helm -f examples/online_serving/chart-helm/values.yaml --set secrets.s3endpoint=http://minio:9000 --set secrets.s3bucketname=testbucket --set secrets.s3accesskeyid=$AWS_ACCESS_KEY_ID --set secrets.s3accesskey=$AWS_SECRET_ACCESS_KEY --set resources.requests.cpu=1 --set resources.requests.memory=4Gi --set resources.limits.cpu=2 --set resources.limits.memory=5Gi --set image.env[0].name=VLLM_CPU_KVCACHE_SPACE --set image.env[1].name=VLLM_LOGGING_LEVEL --set-string image.env[0].value="1" --set-string image.env[1].value="DEBUG" --set-string extraInit.s3modelpath="opt-125m/" --set-string 'resources.limits.nvidia\.com/gpu=0' --set-string 'resources.requests.nvidia\.com/gpu=0' --set-string image.repository="vllm-cpu-env"
-
+    
       - name: curl test
         run: |
           kubectl -n ns-vllm port-forward service/test-vllm-service 8001:80 &
@@ -79,4 +79,4 @@ jobs:
                           "max_tokens": 7,
                           "temperature": 0
                   }'):$CODE"
-          echo "$CODE"
+          echo "$CODE"

.github/workflows/publish.yml

@@ -39,7 +39,7 @@ jobs:
             const script = require('.github/workflows/scripts/create_release.js')
             await script(github, context, core)
 
-  # NOTE(simon): No longer build wheel using GitHub Actions. See buildkite's release workflow. 
+  # NOTE(simon): No longer build wheel using Github Actions. See buildkite's release workflow. 
   # wheel:
   #   name: Build Wheel
   #   runs-on: ${{ matrix.os }}
@@ -50,7 +50,7 @@ jobs:
   #     matrix:
   #         os: ['ubuntu-20.04']
   #         python-version: ['3.9', '3.10', '3.11', '3.12']
-  #         pytorch-version: ['2.4.0']  # Must be the most recent version that meets requirements/cuda.txt.
+  #         pytorch-version: ['2.4.0']  # Must be the most recent version that meets requirements-cuda.txt.
   #         cuda-version: ['11.8', '12.1']
 
   #   steps:

.github/workflows/scripts/build.sh

@@ -9,7 +9,7 @@ PATH=${cuda_home}/bin:$PATH
 LD_LIBRARY_PATH=${cuda_home}/lib64:$LD_LIBRARY_PATH
 
 # Install requirements
-$python_executable -m pip install -r requirements/build.txt -r requirements/cuda.txt
+$python_executable -m pip install -r requirements-build.txt -r requirements-cuda.txt
 
 # Limit the number of parallel jobs to avoid OOM
 export MAX_JOBS=1

.github/workflows/scripts/create_release.js

@@ -1,4 +1,4 @@
-// Uses GitHub's API to create the release and wait for result.
+// Uses Github's API to create the release and wait for result.
 // We use a JS script since github CLI doesn't provide a way to wait for the release's creation and returns immediately.
 
 module.exports = async (github, context, core) => {

.gitignore

@@ -2,7 +2,7 @@
 /vllm/_version.py
 
 # vllm-flash-attn built from source
-vllm/vllm_flash_attn/*
+vllm/vllm_flash_attn/
 
 # Byte-compiled / optimized / DLL files
 __pycache__/
@@ -80,7 +80,6 @@ instance/
 # Sphinx documentation
 docs/_build/
 docs/source/getting_started/examples/
-docs/source/api/vllm
 
 # PyBuilder
 .pybuilder/
@@ -198,11 +197,8 @@ _build/
 hip_compat.h
 
 # Benchmark dataset
-benchmarks/**/*.json
+benchmarks/*.json
 
 # Linting
 actionlint
 shellcheck*/
-
-# Ingore moe/marlin_moe gen code
-csrc/moe/marlin_moe_wna16/kernel_*

.pre-commit-config.yaml

@@ -1,53 +1,48 @@
-default_install_hook_types:
-  - pre-commit
-  - commit-msg
 default_stages:
   - pre-commit # Run locally
   - manual # Run in CI
-exclude: 'vllm/third_party/.*'
 repos:
 - repo: https://github.com/google/yapf
   rev: v0.43.0
   hooks:
   - id: yapf
     args: [--in-place, --verbose]
+    additional_dependencies: [toml] # TODO: Remove when yapf is upgraded
+    exclude: 'vllm/third_party/.*'
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.11.7
+  rev: v0.9.3
   hooks:
   - id: ruff
-    args: [--output-format, github, --fix]
+    args: [--output-format, github]
+    exclude: 'vllm/third_party/.*'
 - repo: https://github.com/codespell-project/codespell
-  rev: v2.4.1
+  rev: v2.4.0
   hooks:
   - id: codespell
-    additional_dependencies: ['tomli']
-    args: ['--toml', 'pyproject.toml']
+    exclude: 'benchmarks/sonnet.txt|(build|tests/(lora/data|models/fixtures|prompts))/.*|vllm/third_party/.*'
 - repo: https://github.com/PyCQA/isort
-  rev: 6.0.1
+  rev: 5.13.2
   hooks:
   - id: isort
+    exclude: 'vllm/third_party/.*'
 - repo: https://github.com/pre-commit/mirrors-clang-format
-  rev: v20.1.3
+  rev: v19.1.7
   hooks:
   - id: clang-format
     exclude: 'csrc/(moe/topk_softmax_kernels.cu|quantization/gguf/(ggml-common.h|dequantize.cuh|vecdotq.cuh|mmq.cuh|mmvq.cuh))|vllm/third_party/.*'
     types_or: [c++, cuda]
     args: [--style=file, --verbose]
 - repo: https://github.com/jackdewinter/pymarkdown
-  rev: v0.9.29
+  rev: v0.9.27
   hooks:
   - id: pymarkdown
     args: [fix]
+    exclude: 'vllm/third_party/.*'
 - repo: https://github.com/rhysd/actionlint
   rev: v1.7.7
   hooks:
   - id: actionlint
-- repo: https://github.com/astral-sh/uv-pre-commit
-  rev: 0.6.17
-  hooks:
-    - id: pip-compile
-      args: [requirements/test.in, -o, requirements/test.txt, --index-strategy, unsafe-best-match, --torch-backend, cu128]
-      files: ^requirements/test\.(in|txt)$
+    exclude: 'vllm/third_party/.*'
 - repo: local
   hooks:
   - id: mypy-local
@@ -55,8 +50,9 @@ repos:
     entry: tools/mypy.sh 0 "local"
     language: python
     types: [python]
-    additional_dependencies: &mypy_deps [mypy==1.11.1, types-cachetools, types-setuptools, types-PyYAML, types-requests]
+    additional_dependencies: &mypy_deps [mypy==1.11.1, types-setuptools, types-PyYAML, types-requests]
     stages: [pre-commit] # Don't run in CI
+    exclude: 'vllm/third_party/.*'
   - id: mypy-3.9 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
     name: Run mypy for Python 3.9
     entry: tools/mypy.sh 1 "3.9"
@@ -64,6 +60,7 @@ repos:
     types: [python]
     additional_dependencies: *mypy_deps
     stages: [manual] # Only run in CI
+    exclude: 'vllm/third_party/.*'
   - id: mypy-3.10 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
     name: Run mypy for Python 3.10
     entry: tools/mypy.sh 1 "3.10"
@@ -71,6 +68,7 @@ repos:
     types: [python]
     additional_dependencies: *mypy_deps
     stages: [manual] # Only run in CI
+    exclude: 'vllm/third_party/.*'
   - id: mypy-3.11 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
     name: Run mypy for Python 3.11
     entry: tools/mypy.sh 1 "3.11"
@@ -78,6 +76,7 @@ repos:
     types: [python]
     additional_dependencies: *mypy_deps
     stages: [manual] # Only run in CI
+    exclude: 'vllm/third_party/.*'
   - id: mypy-3.12 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
     name: Run mypy for Python 3.12
     entry: tools/mypy.sh 1 "3.12"
@@ -85,33 +84,38 @@ repos:
     types: [python]
     additional_dependencies: *mypy_deps
     stages: [manual] # Only run in CI
+    exclude: 'vllm/third_party/.*'
   - id: shellcheck
     name: Lint shell scripts
     entry: tools/shellcheck.sh
     language: script
     types: [shell]
+    exclude: 'vllm/third_party/.*'
   - id: png-lint
     name: Lint PNG exports from excalidraw
     entry: tools/png-lint.sh
     language: script
     types: [png]
+    exclude: 'vllm/third_party/.*'
   - id: signoff-commit
     name: Sign-off Commit
     entry: bash
     args:
       - -c
       - |
-        if ! grep -q "^Signed-off-by: $(git config user.name) <$(git config user.email)>" "$(git rev-parse --git-path COMMIT_EDITMSG)"; then
-          printf "\nSigned-off-by: $(git config user.name) <$(git config user.email)>\n" >> "$(git rev-parse --git-path COMMIT_EDITMSG)"
+        if ! grep -q "^Signed-off-by: $(git config user.name) <$(git config user.email)>" .git/COMMIT_EDITMSG; then
+          printf "\nSigned-off-by: $(git config user.name) <$(git config user.email)>\n" >> .git/COMMIT_EDITMSG
         fi
     language: system
     verbose: true
     stages: [commit-msg]
+    exclude: 'vllm/third_party/.*'
   - id: check-spdx-header
     name: Check SPDX headers
     entry: python tools/check_spdx_header.py
     language: python
     types: [python]
+    exclude: 'vllm/third_party/.*'
   - id: check-filenames
     name: Check for spaces in all filenames
     entry: bash
@@ -121,10 +125,7 @@ repos:
     language: system
     always_run: true
     pass_filenames: false
-  - id: update-dockerfile-graph
-    name: Update Dockerfile dependency graph
-    entry: tools/update-dockerfile-graph.sh
-    language: script
+    exclude: 'vllm/third_party/.*'
   # Keep `suggestion` last
   - id: suggestion
     name: Suggestion
@@ -132,4 +133,5 @@ repos:
     language: system
     verbose: true
     pass_filenames: false
+    exclude: 'vllm/third_party/.*'
   # Insert new entries above the `suggestion` entry

.readthedocs.yaml

@@ -18,4 +18,4 @@ formats: []
 # Optionally declare the Python requirements required to build your docs
 python:
   install:
-    - requirements: requirements/docs.txt
+    - requirements: docs/requirements-docs.txt

CMakeLists.txt

@@ -15,6 +15,7 @@ project(vllm_extensions LANGUAGES CXX)
 
 # CUDA by default, can be overridden by using -DVLLM_TARGET_DEVICE=... (used by setup.py)
 set(VLLM_TARGET_DEVICE "cuda" CACHE STRING "Target device backend for vLLM")
+
 message(STATUS "Build type: ${CMAKE_BUILD_TYPE}")
 message(STATUS "Target device: ${VLLM_TARGET_DEVICE}")
 
@@ -30,10 +31,10 @@ set(ignoreMe "${VLLM_PYTHON_PATH}")
 set(PYTHON_SUPPORTED_VERSIONS "3.9" "3.10" "3.11" "3.12")
 
 # Supported NVIDIA architectures.
-set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0")
+set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0")
 
 # Supported AMD GPU architectures.
-set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1101;gfx1200;gfx1201")
+set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101")
 
 #
 # Supported/expected torch versions for CUDA/ROCm.
@@ -43,10 +44,10 @@ set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1
 #
 # Note: the CUDA torch version is derived from pyproject.toml and various
 # requirements.txt files and should be kept consistent.  The ROCm torch
-# versions are derived from docker/Dockerfile.rocm
+# versions are derived from Dockerfile.rocm
 #
-set(TORCH_SUPPORTED_VERSION_CUDA "2.7.0")
-set(TORCH_SUPPORTED_VERSION_ROCM "2.7.0")
+set(TORCH_SUPPORTED_VERSION_CUDA "2.5.1")
+set(TORCH_SUPPORTED_VERSION_ROCM "2.5.1")
 
 #
 # Try to find python package with an executable that exactly matches
@@ -173,25 +174,6 @@ include(FetchContent)
 file(MAKE_DIRECTORY ${FETCHCONTENT_BASE_DIR}) # Ensure the directory exists
 message(STATUS "FetchContent base directory: ${FETCHCONTENT_BASE_DIR}")
 
-#
-# Set rocm version dev int.
-#
-if(VLLM_GPU_LANG STREQUAL "HIP")
-  #
-  # Overriding the default -O set up by cmake, adding ggdb3 for the most verbose devug info
-  #
-  set(CMAKE_${VLLM_GPU_LANG}_FLAGS_DEBUG "${CMAKE_${VLLM_GPU_LANG}_FLAGS_DEBUG} -O0 -ggdb3")
-  set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -O0 -ggdb3")
-
-
-  #
-  # Certain HIP functions are marked as [[nodiscard]], yet vllm ignores the result which generates
-  # a lot of warnings that always mask real issues. Suppressing until this is properly addressed.
-  #
-  set(CMAKE_${VLLM_GPU_LANG}_FLAGS "${CMAKE_${VLLM_GPU_LANG}_FLAGS} -Wno-unused-result")
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-unused-result")
-endif()
-
 #
 # Define other extension targets
 #
@@ -229,28 +211,24 @@ set(VLLM_EXT_SRC
   "csrc/cache_kernels.cu"
   "csrc/attention/paged_attention_v1.cu"
   "csrc/attention/paged_attention_v2.cu"
-  "csrc/attention/merge_attn_states.cu"
   "csrc/pos_encoding_kernels.cu"
   "csrc/activation_kernels.cu"
   "csrc/layernorm_kernels.cu"
   "csrc/layernorm_quant_kernels.cu"
-  "csrc/cuda_view.cu"
   "csrc/quantization/gptq/q_gemm.cu"
   "csrc/quantization/compressed_tensors/int8_quant_kernels.cu"
   "csrc/quantization/fp8/common.cu"
   "csrc/quantization/fused_kernels/fused_layernorm_dynamic_per_token_quant.cu"
   "csrc/quantization/gguf/gguf_kernel.cu"
-  "csrc/quantization/activation_kernels.cu"
   "csrc/cuda_utils_kernels.cu"
   "csrc/prepare_inputs/advance_step.cu"
-  "csrc/custom_all_reduce.cu"
   "csrc/torch_bindings.cpp")
 
 if(VLLM_GPU_LANG STREQUAL "CUDA")
   SET(CUTLASS_ENABLE_HEADERS_ONLY ON CACHE BOOL "Enable only the header library")
 
-  # Set CUTLASS_REVISION. Used for FetchContent. Also fixes some bogus messages when building.
-  set(CUTLASS_REVISION "v3.9.2" CACHE STRING "CUTLASS revision to use")
+  # Set CUTLASS_REVISION manually -- its revision detection doesn't work in this case.
+  set(CUTLASS_REVISION "v3.6.0" CACHE STRING "CUTLASS revision to use")
 
   # Use the specified CUTLASS source directory for compilation if VLLM_CUTLASS_SRC_DIR is provided
   if (DEFINED ENV{VLLM_CUTLASS_SRC_DIR})
@@ -267,8 +245,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     FetchContent_Declare(
         cutlass
         GIT_REPOSITORY https://github.com/nvidia/cutlass.git
-        # Please keep this in sync with CUTLASS_REVISION line above.
-        GIT_TAG ${CUTLASS_REVISION}
+        GIT_TAG v3.7.0
         GIT_PROGRESS TRUE
 
         # Speed up CUTLASS download by retrieving only the specified GIT_TAG instead of the history.
@@ -284,13 +261,12 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     "csrc/mamba/causal_conv1d/causal_conv1d.cu"
     "csrc/quantization/aqlm/gemm_kernels.cu"
     "csrc/quantization/awq/gemm_kernels.cu"
+    "csrc/custom_all_reduce.cu"
     "csrc/permute_cols.cu"
     "csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu"
-    "csrc/quantization/fp4/nvfp4_quant_entry.cu"
-    "csrc/quantization/fp4/nvfp4_scaled_mm_entry.cu"
     "csrc/sparse/cutlass/sparse_scaled_mm_entry.cu"
-    "csrc/cutlass_extensions/common.cpp"
-    "csrc/attention/mla/cutlass_mla_entry.cu")
+    "csrc/sparse/cutlass/sparse_compressor_entry.cu"
+    "csrc/cutlass_extensions/common.cpp")
 
   set_gencode_flags_for_srcs(
     SRCS "${VLLM_EXT_SRC}"
@@ -299,54 +275,10 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
   # Only build Marlin kernels if we are building for at least some compatible archs.
   # Keep building Marlin for 9.0 as there are some group sizes and shapes that
   # are not supported by Machete yet.
-  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0" "${CUDA_ARCHS}")
+  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.6;8.7;8.9;9.0" "${CUDA_ARCHS}")
   if (MARLIN_ARCHS)
-
-    #
-    # For the Marlin kernels we automatically generate sources for various
-    # preselected input type pairs and schedules.
-    # Generate sources:
-    set(MARLIN_GEN_SCRIPT
-      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/gptq_marlin/generate_kernels.py)
-    file(MD5 ${MARLIN_GEN_SCRIPT} MARLIN_GEN_SCRIPT_HASH)
-
-    message(STATUS "Marlin generation script hash: ${MARLIN_GEN_SCRIPT_HASH}")
-    message(STATUS "Last run Marlin generate script hash: $CACHE{MARLIN_GEN_SCRIPT_HASH}")
-
-    if (NOT DEFINED CACHE{MARLIN_GEN_SCRIPT_HASH}
-        OR NOT $CACHE{MARLIN_GEN_SCRIPT_HASH} STREQUAL ${MARLIN_GEN_SCRIPT_HASH})
-      execute_process(
-        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=$PYTHONPATH
-          ${Python_EXECUTABLE} ${MARLIN_GEN_SCRIPT}
-        RESULT_VARIABLE marlin_generation_result
-        OUTPUT_VARIABLE marlin_generation_result
-        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log
-        ERROR_FILE ${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log
-      )
-
-      if (NOT marlin_generation_result EQUAL 0)
-        message(FATAL_ERROR "Marlin generation failed."
-                            " Result: \"${marlin_generation_result}\""
-                            "\nCheck the log for details: "
-                            "${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log")
-      else()
-        set(MARLIN_GEN_SCRIPT_HASH ${MARLIN_GEN_SCRIPT_HASH}
-            CACHE STRING "Last run Marlin generate script hash" FORCE)
-        message(STATUS "Marlin generation completed successfully.")
-      endif()
-    else()
-      message(STATUS "Marlin generation script has not changed, skipping generation.")
-    endif()
-
-    file(GLOB MARLIN_TEMPLATE_KERNEL_SRC "csrc/quantization/gptq_marlin/kernel_*.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${MARLIN_TEMPLATE_KERNEL_SRC}"
-      CUDA_ARCHS "${MARLIN_ARCHS}")
-
-    list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_KERNEL_SRC})
-
     set(MARLIN_SRCS
+       "csrc/quantization/fp8/fp8_marlin.cu"
        "csrc/quantization/marlin/dense/marlin_cuda_kernel.cu"
        "csrc/quantization/marlin/sparse/marlin_24_cuda_kernel.cu"
        "csrc/quantization/marlin/qqq/marlin_qqq_gemm_kernel.cu"
@@ -363,88 +295,43 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
                    " in CUDA target architectures")
   endif()
 
-  # Only build AllSpark kernels if we are building for at least some compatible archs.
-  cuda_archs_loose_intersection(ALLSPARK_ARCHS "8.0;8.6;8.7;8.9" "${CUDA_ARCHS}")
-  if (ALLSPARK_ARCHS)
-    set(ALLSPARK_SRCS
-       "csrc/quantization/gptq_allspark/allspark_repack.cu"
-       "csrc/quantization/gptq_allspark/allspark_qgemm_w8a16.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${ALLSPARK_SRCS}"
-      CUDA_ARCHS "${ALLSPARK_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${ALLSPARK_SRCS}")
-    message(STATUS "Building AllSpark kernels for archs: ${ALLSPARK_ARCHS}")
-  else()
-    message(STATUS "Not building AllSpark kernels as no compatible archs found"
-                   " in CUDA target architectures")
-  endif()
-
-
-  set(SCALED_MM_3X_ARCHS)
   # The cutlass_scaled_mm kernels for Hopper (c3x, i.e. CUTLASS 3.x) require
-  # CUDA 12.0 or later
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_ARCHS)
-    set(SRCS
-       "csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm90.cu"
+  # CUDA 12.0 or later (and only work on Hopper, 9.0a for now).
+  cuda_archs_loose_intersection(SCALED_MM_3X_ARCHS "9.0a" "${CUDA_ARCHS}")
+  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_3X_ARCHS)
+    set(SRCS 
+       "csrc/quantization/cutlass_w8a8/scaled_mm_c3x.cu"
        "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_fp8.cu"
        "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_int8.cu"
        "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_azp_sm90_int8.cu"
        "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm90_fp8.cu")
     set_gencode_flags_for_srcs(
       SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
+      CUDA_ARCHS "${SCALED_MM_3X_ARCHS}")
     list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_SM90=1")
-    # Let scaled_mm_c2x know it doesn't need to build these arches
-    list(APPEND SCALED_MM_3X_ARCHS "${SCALED_MM_ARCHS}")
-    message(STATUS "Building scaled_mm_c3x_sm90 for archs: ${SCALED_MM_ARCHS}")
+    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_C3X=1")
+    message(STATUS "Building scaled_mm_c3x for archs: ${SCALED_MM_3X_ARCHS}")
   else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building scaled_mm_c3x_sm90 as CUDA Compiler version is "
+    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_3X_ARCHS)
+      message(STATUS "Not building scaled_mm_c3x as CUDA Compiler version is "
                      "not >= 12.0, we recommend upgrading to CUDA 12.0 or "
                      "later if you intend on running FP8 quantized models on "
                      "Hopper.")
     else()
-      message(STATUS "Not building scaled_mm_c3x_sm90 as no compatible archs found "
+      message(STATUS "Not building scaled_mm_c3x as no compatible archs found "
                      "in CUDA target architectures")
     endif()
-  endif()
 
-  # The cutlass_scaled_mm kernels for Blackwell (c3x, i.e. CUTLASS 3.x) require
-  # CUDA 12.8 or later
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;12.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND SCALED_MM_ARCHS)
-    set(SRCS
-      "csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm100.cu"
-      "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm100_fp8.cu"
-      "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm100_fp8.cu"
-    )
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_SM100=1")
-    # Let scaled_mm_c2x know it doesn't need to build these arches
-    list(APPEND SCALED_MM_3X_ARCHS "${SCALED_MM_ARCHS}")
-    message(STATUS "Building scaled_mm_c3x_sm100 for archs: ${SCALED_MM_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building scaled_mm_c3x_sm100 as CUDA Compiler version is "
-                     "not >= 12.8, we recommend upgrading to CUDA 12.8 or "
-                     "later if you intend on running FP8 quantized models on "
-                     "Blackwell.")
-    else()
-      message(STATUS "Not building scaled_mm_c3x_100 as no compatible archs found "
-                     "in CUDA target architectures")
-    endif()
+    # clear SCALED_MM_3X_ARCHS so the scaled_mm_c2x kernels know we didn't
+    # build any 3x kernels
+    set(SCALED_MM_3X_ARCHS)
   endif()
 
   #
   # For the cutlass_scaled_mm kernels we want to build the c2x (CUTLASS 2.x)
   # kernels for the remaining archs that are not already built for 3x.
   cuda_archs_loose_intersection(SCALED_MM_2X_ARCHS
-    "7.5;8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0" "${CUDA_ARCHS}")
+    "7.5;8.0;8.6;8.7;8.9;9.0" "${CUDA_ARCHS}")
   # subtract out the archs that are already built for 3x
   list(REMOVE_ITEM SCALED_MM_2X_ARCHS ${SCALED_MM_3X_ARCHS})
   if (SCALED_MM_2X_ARCHS)
@@ -469,18 +356,18 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
   # 2:4 Sparse Kernels
 
   # The 2:4 sparse kernels cutlass_scaled_sparse_mm and cutlass_compressor
-  # require CUDA 12.2 or later (and only work on Hopper).
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_ARCHS)
-    set(SRCS "csrc/sparse/cutlass/sparse_scaled_mm_c3x.cu")
+  # require CUDA 12.2 or later (and only work on Hopper, 9.0a for now).
+  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_3X_ARCHS)
+    set(SRCS "csrc/sparse/cutlass/sparse_compressor_c3x.cu"
+             "csrc/sparse/cutlass/sparse_scaled_mm_c3x.cu")
     set_gencode_flags_for_srcs(
       SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
+      CUDA_ARCHS "${SCALED_MM_3X_ARCHS}")
     list(APPEND VLLM_EXT_SRC "${SRCS}")
     list(APPEND VLLM_GPU_FLAGS "-DENABLE_SPARSE_SCALED_MM_C3X=1")
-    message(STATUS "Building sparse_scaled_mm_c3x for archs: ${SCALED_MM_ARCHS}")
+    message(STATUS "Building sparse_scaled_mm_c3x for archs: ${SCALED_MM_3X_ARCHS}")
   else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_ARCHS)
+    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_3X_ARCHS)
       message(STATUS "Not building sparse_scaled_mm_c3x kernels as CUDA Compiler version is "
                      "not >= 12.2, we recommend upgrading to CUDA 12.2 or later "
                      "if you intend on running FP8 sparse quantized models on Hopper.")
@@ -490,69 +377,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     endif()
   endif()
 
-  # FP4 Archs and flags
-  cuda_archs_loose_intersection(FP4_ARCHS "10.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND FP4_ARCHS)
-    set(SRCS
-      "csrc/quantization/fp4/nvfp4_quant_kernels.cu"
-      "csrc/quantization/fp4/nvfp4_scaled_mm_kernels.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${FP4_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_NVFP4=1")
-    message(STATUS "Building NVFP4 for archs: ${FP4_ARCHS}")
-  else()
-    message(STATUS "Not building NVFP4 as no compatible archs were found.")
-    # clear FP4_ARCHS
-    set(FP4_ARCHS)
-  endif()
-
-  # CUTLASS MLA Archs and flags
-  cuda_archs_loose_intersection(MLA_ARCHS "10.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND MLA_ARCHS)
-    set(SRCS
-      "csrc/attention/mla/cutlass_mla_kernels.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${MLA_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MLA=1")
-    # Add MLA-specific include directories only to MLA source files
-    set_source_files_properties(${SRCS}
-      PROPERTIES INCLUDE_DIRECTORIES "${CUTLASS_DIR}/examples/77_blackwell_fmha;${CUTLASS_DIR}/examples/common")
-    message(STATUS "Building CUTLASS MLA for archs: ${MLA_ARCHS}")
-  else()
-    message(STATUS "Not building CUTLASS MLA as no compatible archs were found.")
-    # clear MLA_ARCHS
-    set(MLA_ARCHS)
-  endif()
-
-  # CUTLASS MoE kernels
-
-  # The MoE kernel cutlass_moe_mm requires CUDA 12.3 or later (and only works
-  # on Hopper). get_cutlass_moe_mm_data should only be compiled if it's possible
-  # to compile MoE kernels that use its output.
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
-    set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu"
-             "csrc/quantization/cutlass_w8a8/moe/moe_data.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MOE_SM90=1")
-    message(STATUS "Building grouped_mm_c3x for archs: ${SCALED_MM_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building grouped_mm_c3x kernels as CUDA Compiler version is "
-                     "not >= 12.3, we recommend upgrading to CUDA 12.3 or later "
-                     "if you intend on running FP8 quantized MoE models on Hopper.")
-    else()
-      message(STATUS "Not building grouped_mm_c3x as no compatible archs found "
-                     "in CUDA target architectures")
-    endif()
-  endif()
 
   #
   # Machete kernels
@@ -634,8 +458,7 @@ define_gpu_extension_target(
   SOURCES ${VLLM_EXT_SRC}
   COMPILE_FLAGS ${VLLM_GPU_FLAGS}
   ARCHITECTURES ${VLLM_GPU_ARCHES}
-  INCLUDE_DIRECTORIES ${CUTLASS_INCLUDE_DIR}
-  INCLUDE_DIRECTORIES ${CUTLASS_TOOLS_UTIL_INCLUDE_DIR}
+  INCLUDE_DIRECTORIES ${CUTLASS_INCLUDE_DIR};${CUTLASS_TOOLS_UTIL_INCLUDE_DIR}
   USE_SABI 3
   WITH_SOABI)
 
@@ -654,70 +477,28 @@ set(VLLM_MOE_EXT_SRC
   "csrc/moe/moe_align_sum_kernels.cu"
   "csrc/moe/topk_softmax_kernels.cu")
 
-if(VLLM_GPU_LANG STREQUAL "CUDA")
-  list(APPEND VLLM_MOE_EXT_SRC "csrc/moe/moe_wna16.cu")
-endif()
-
 set_gencode_flags_for_srcs(
   SRCS "${VLLM_MOE_EXT_SRC}"
   CUDA_ARCHS "${CUDA_ARCHS}")
 
 if(VLLM_GPU_LANG STREQUAL "CUDA")
-  set(VLLM_MOE_WNA16_SRC
-    "csrc/moe/moe_wna16.cu")
-
-  set_gencode_flags_for_srcs(
-    SRCS "${VLLM_MOE_WNA16_SRC}"
-    CUDA_ARCHS "${CUDA_ARCHS}")
-
-  list(APPEND VLLM_MOE_EXT_SRC "${VLLM_MOE_WNA16_SRC}")
-  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0" "${CUDA_ARCHS}")
+  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.6;8.7;8.9;9.0" "${CUDA_ARCHS}")
   if (MARLIN_MOE_ARCHS)
+    set(MARLIN_MOE_SRC
+        "csrc/moe/marlin_kernels/marlin_moe_kernel.h"
+        "csrc/moe/marlin_kernels/marlin_moe_kernel_ku4b8.h"
+        "csrc/moe/marlin_kernels/marlin_moe_kernel_ku4b8.cu"
+        "csrc/moe/marlin_kernels/marlin_moe_kernel_ku8b128.h"
+        "csrc/moe/marlin_kernels/marlin_moe_kernel_ku8b128.cu"
+        "csrc/moe/marlin_kernels/marlin_moe_kernel_ku4.h"
+        "csrc/moe/marlin_kernels/marlin_moe_kernel_ku4.cu"
+        "csrc/moe/marlin_moe_ops.cu")
 
-    #
-    # For the Marlin MOE kernels we automatically generate sources for various
-    # preselected input type pairs and schedules.
-    # Generate sources:
-    set(MOE_MARLIN_GEN_SCRIPT
-      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/moe/marlin_moe_wna16/generate_kernels.py)
-    file(MD5 ${MOE_MARLIN_GEN_SCRIPT} MOE_MARLIN_GEN_SCRIPT_HASH)
-
-    message(STATUS "Marlin MOE generation script hash: ${MOE_MARLIN_GEN_SCRIPT_HASH}")
-    message(STATUS "Last run Marlin MOE generate script hash: $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH}")
-
-    if (NOT DEFINED CACHE{MOE_MARLIN_GEN_SCRIPT_HASH}
-        OR NOT $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH} STREQUAL ${MOE_MARLIN_GEN_SCRIPT_HASH})
-      execute_process(
-        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=$PYTHONPATH
-          ${Python_EXECUTABLE} ${MOE_MARLIN_GEN_SCRIPT}
-        RESULT_VARIABLE moe_marlin_generation_result
-        OUTPUT_VARIABLE moe_marlin_generation_output
-        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log
-        ERROR_FILE ${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log
-      )
-
-      if (NOT moe_marlin_generation_result EQUAL 0)
-        message(FATAL_ERROR "Marlin MOE generation failed."
-                            " Result: \"${moe_marlin_generation_result}\""
-                            "\nCheck the log for details: "
-                            "${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log")
-      else()
-        set(MOE_MARLIN_GEN_SCRIPT_HASH ${MOE_MARLIN_GEN_SCRIPT_HASH}
-            CACHE STRING "Last run Marlin MOE generate script hash" FORCE)
-        message(STATUS "Marlin MOE generation completed successfully.")
-      endif()
-    else()
-      message(STATUS "Marlin MOE generation script has not changed, skipping generation.")
-    endif()
-
-    file(GLOB MOE_WNAA16_MARLIN_SRC "csrc/moe/marlin_moe_wna16/*.cu")
     set_gencode_flags_for_srcs(
-      SRCS "${MOE_WNAA16_MARLIN_SRC}"
+      SRCS "${MARLIN_MOE_SRC}"
       CUDA_ARCHS "${MARLIN_MOE_ARCHS}")
 
-    list(APPEND VLLM_MOE_EXT_SRC ${MOE_WNAA16_MARLIN_SRC})
-
+    list(APPEND VLLM_MOE_EXT_SRC "${MARLIN_MOE_SRC}")
     message(STATUS "Building Marlin MOE kernels for archs: ${MARLIN_MOE_ARCHS}")
   else()
     message(STATUS "Not building Marlin MOE kernels as no compatible archs found"
@@ -725,17 +506,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
   endif()
 endif()
 
-if(VLLM_GPU_LANG STREQUAL "CUDA")
-  set(MOE_PERMUTE_SRC
-      "csrc/moe/permute_unpermute_kernels/moe_permute_unpermute_kernel.cu"
-      "csrc/moe/moe_permute_unpermute_op.cu")
-
-  set_gencode_flags_for_srcs(
-    SRCS "${MARLIN_PERMUTE_SRC}"
-    CUDA_ARCHS "${MOE_PERMUTE_ARCHS}")
-
-  list(APPEND VLLM_MOE_EXT_SRC "${MOE_PERMUTE_SRC}")
-endif()
 message(STATUS "Enabling moe extension.")
 define_gpu_extension_target(
   _moe_C
@@ -744,8 +514,6 @@ define_gpu_extension_target(
   SOURCES ${VLLM_MOE_EXT_SRC}
   COMPILE_FLAGS ${VLLM_GPU_FLAGS}
   ARCHITECTURES ${VLLM_GPU_ARCHES}
-  INCLUDE_DIRECTORIES ${CUTLASS_INCLUDE_DIR}
-  INCLUDE_DIRECTORIES ${CUTLASS_TOOLS_UTIL_INCLUDE_DIR}
   USE_SABI 3
   WITH_SOABI)
 
@@ -755,7 +523,6 @@ if(VLLM_GPU_LANG STREQUAL "HIP")
   #
   set(VLLM_ROCM_EXT_SRC
     "csrc/rocm/torch_bindings.cpp"
-    "csrc/rocm/skinny_gemms.cu"
     "csrc/rocm/attention.cu")
 
   define_gpu_extension_target(
@@ -769,8 +536,77 @@ if(VLLM_GPU_LANG STREQUAL "HIP")
     WITH_SOABI)
 endif()
 
-# For CUDA we also build and ship some external projects.
-if (VLLM_GPU_LANG STREQUAL "CUDA")
-    include(cmake/external_projects/flashmla.cmake)
-    include(cmake/external_projects/vllm_flash_attn.cmake)
+# vllm-flash-attn currently only supported on CUDA
+if (NOT VLLM_GPU_LANG STREQUAL "CUDA")
+  return()
 endif ()
+
+# vLLM flash attention requires VLLM_GPU_ARCHES to contain the set of target
+# arches in the CMake syntax (75-real, 89-virtual, etc), since we clear the
+# arches in the CUDA case (and instead set the gencodes on a per file basis)
+# we need to manually set VLLM_GPU_ARCHES here.
+if(VLLM_GPU_LANG STREQUAL "CUDA")
+  foreach(_ARCH ${CUDA_ARCHS})
+    string(REPLACE "." "" _ARCH "${_ARCH}")
+    list(APPEND VLLM_GPU_ARCHES "${_ARCH}-real")
+  endforeach()
+endif()
+
+#
+# Build vLLM flash attention from source
+#
+# IMPORTANT: This has to be the last thing we do, because vllm-flash-attn uses the same macros/functions as vLLM.
+# Because functions all belong to the global scope, vllm-flash-attn's functions overwrite vLLMs.
+# They should be identical but if they aren't, this is a massive footgun.
+#
+# The vllm-flash-attn install rules are nested under vllm to make sure the library gets installed in the correct place.
+# To only install vllm-flash-attn, use --component _vllm_fa2_C (for FA2) or --component _vllm_fa3_C (for FA3).
+# If no component is specified, vllm-flash-attn is still installed.
+
+# If VLLM_FLASH_ATTN_SRC_DIR is set, vllm-flash-attn is installed from that directory instead of downloading.
+# This is to enable local development of vllm-flash-attn within vLLM.
+# It can be set as an environment variable or passed as a cmake argument.
+# The environment variable takes precedence.
+if (DEFINED ENV{VLLM_FLASH_ATTN_SRC_DIR})
+  set(VLLM_FLASH_ATTN_SRC_DIR $ENV{VLLM_FLASH_ATTN_SRC_DIR})
+endif()
+
+if(VLLM_FLASH_ATTN_SRC_DIR)
+  FetchContent_Declare(
+          vllm-flash-attn SOURCE_DIR 
+          ${VLLM_FLASH_ATTN_SRC_DIR}
+          BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn
+  )
+else()
+  FetchContent_Declare(
+          vllm-flash-attn
+          GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
+          GIT_TAG 720c94869cf2e0ff5a706e9c7f1dce0939686ade
+          GIT_PROGRESS TRUE
+          # Don't share the vllm-flash-attn build between build types
+          BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn
+  )
+endif()
+
+
+# Fetch the vllm-flash-attn library
+FetchContent_MakeAvailable(vllm-flash-attn)
+message(STATUS "vllm-flash-attn is available at ${vllm-flash-attn_SOURCE_DIR}")
+
+# Copy over the vllm-flash-attn python files (duplicated for fa2 and fa3, in
+# case only one is built, in the case both are built redundant work is done)
+install(
+  DIRECTORY ${vllm-flash-attn_SOURCE_DIR}/vllm_flash_attn/
+  DESTINATION vllm_flash_attn
+  COMPONENT _vllm_fa2_C
+  FILES_MATCHING PATTERN "*.py"
+)
+
+install(
+  DIRECTORY ${vllm-flash-attn_SOURCE_DIR}/vllm_flash_attn/
+  DESTINATION vllm_flash_attn
+  COMPONENT _vllm_fa3_C
+  FILES_MATCHING PATTERN "*.py"
+)
+
+# Nothing after vllm-flash-attn, see comment about macros above

Dockerfile

@@ -5,11 +5,11 @@
 # docs/source/contributing/dockerfile/dockerfile.md and
 # docs/source/assets/contributing/dockerfile-stages-dependency.png
 
-ARG CUDA_VERSION=12.8.1
+ARG CUDA_VERSION=12.4.1
 #################### BASE BUILD IMAGE ####################
 # prepare basic build environment
 FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu20.04 AS base
-ARG CUDA_VERSION=12.8.1
+ARG CUDA_VERSION=12.4.1
 ARG PYTHON_VERSION=3.12
 ARG TARGETPLATFORM
 ENV DEBIAN_FRONTEND=noninteractive
@@ -19,10 +19,7 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
     && echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \
     && apt-get update -y \
     && apt-get install -y ccache software-properties-common git curl sudo \
-    && for i in 1 2 3; do \
-        add-apt-repository -y ppa:deadsnakes/ppa && break || \
-        { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
-    done \
+    && add-apt-repository ppa:deadsnakes/ppa \
     && apt-get update -y \
     && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
     && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
@@ -30,14 +27,6 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
     && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
     && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \
     && python3 --version && python3 -m pip --version
-# Install uv for faster pip installs
-RUN --mount=type=cache,target=/root/.cache/uv \
-    python3 -m pip install uv
-
-# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
-# Reference: https://github.com/astral-sh/uv/pull/1694
-ENV UV_HTTP_TIMEOUT=500
-ENV UV_INDEX_STRATEGY="unsafe-best-match"
 
 # Upgrade to GCC 10 to avoid https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92519
 # as it was causing spam when compiling the CUTLASS kernels
@@ -61,17 +50,15 @@ WORKDIR /workspace
 # we need to install torch and torchvision from the nightly builds first,
 # pytorch will not appear as a vLLM dependency in all of the following steps
 # after this step
-RUN --mount=type=cache,target=/root/.cache/uv \
+RUN --mount=type=cache,target=/root/.cache/pip \
     if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
-        uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319";  \
-        uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 --pre pytorch_triton==3.3.0+gitab727c40; \
+        python3 -m pip install --index-url https://download.pytorch.org/whl/nightly/cu126 "torch==2.7.0.dev20250121+cu126" "torchvision==0.22.0.dev20250121";  \
     fi
 
-COPY requirements/common.txt requirements/common.txt
-COPY requirements/cuda.txt requirements/cuda.txt
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system -r requirements/cuda.txt \
-    --extra-index-url https://download.pytorch.org/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+COPY requirements-common.txt requirements-common.txt
+COPY requirements-cuda.txt requirements-cuda.txt
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install -r requirements-cuda.txt
 
 # cuda arch list used by torch
 # can be useful for both `dev` and `test`
@@ -89,21 +76,15 @@ FROM base AS build
 ARG TARGETPLATFORM
 
 # install build dependencies
-COPY requirements/build.txt requirements/build.txt
+COPY requirements-build.txt requirements-build.txt
 
-# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
-# Reference: https://github.com/astral-sh/uv/pull/1694
-ENV UV_HTTP_TIMEOUT=500
-ENV UV_INDEX_STRATEGY="unsafe-best-match"
-
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system -r requirements/build.txt \
-    --extra-index-url https://download.pytorch.org/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install -r requirements-build.txt
 
 COPY . .
 ARG GIT_REPO_CHECK=0
 RUN --mount=type=bind,source=.git,target=.git \
-    if [ "$GIT_REPO_CHECK" != "0" ]; then bash tools/check_repo.sh ; fi
+    if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh ; fi
 
 # max jobs used by Ninja to build extensions
 ARG max_jobs=2
@@ -117,7 +98,7 @@ ARG SCCACHE_BUCKET_NAME=vllm-build-sccache
 ARG SCCACHE_REGION_NAME=us-west-2
 ARG SCCACHE_S3_NO_CREDENTIALS=0
 # if USE_SCCACHE is set, use sccache to speed up compilation
-RUN --mount=type=cache,target=/root/.cache/uv \
+RUN --mount=type=cache,target=/root/.cache/pip \
     --mount=type=bind,source=.git,target=.git \
     if [ "$USE_SCCACHE" = "1" ]; then \
         echo "Installing sccache..." \
@@ -137,12 +118,9 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 
 ENV CCACHE_DIR=/root/.cache/ccache
 RUN --mount=type=cache,target=/root/.cache/ccache \
-    --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=cache,target=/root/.cache/pip \
     --mount=type=bind,source=.git,target=.git  \
     if [ "$USE_SCCACHE" != "1" ]; then \
-        # Clean any existing CMake artifacts
-        rm -rf .deps && \
-        mkdir -p .deps && \
         python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38; \
     fi
 
@@ -162,28 +140,18 @@ RUN if [ "$RUN_WHEEL_CHECK" = "true" ]; then \
 #################### DEV IMAGE ####################
 FROM base as dev
 
-# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
-# Reference: https://github.com/astral-sh/uv/pull/1694
-ENV UV_HTTP_TIMEOUT=500
-ENV UV_INDEX_STRATEGY="unsafe-best-match"
-
-# Workaround for #17068
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system --no-build-isolation "git+https://github.com/state-spaces/mamba@v2.2.4"
-
-COPY requirements/lint.txt requirements/lint.txt
-COPY requirements/test.txt requirements/test.txt
-COPY requirements/dev.txt requirements/dev.txt
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system -r requirements/dev.txt \
-    --extra-index-url https://download.pytorch.org/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+COPY requirements-lint.txt requirements-lint.txt
+COPY requirements-test.txt requirements-test.txt
+COPY requirements-dev.txt requirements-dev.txt
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install -r requirements-dev.txt
 #################### DEV IMAGE ####################
 
 #################### vLLM installation IMAGE ####################
 # image with vLLM installed
 # TODO: Restore to base image after FlashInfer AOT wheel fixed
 FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 AS vllm-base
-ARG CUDA_VERSION=12.8.1
+ARG CUDA_VERSION=12.4.1
 ARG PYTHON_VERSION=3.12
 WORKDIR /vllm-workspace
 ENV DEBIAN_FRONTEND=noninteractive
@@ -198,10 +166,7 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
     && apt-get update -y \
     && apt-get install -y ccache software-properties-common git curl wget sudo vim python3-pip \
     && apt-get install -y ffmpeg libsm6 libxext6 libgl1 \
-    && for i in 1 2 3; do \
-        add-apt-repository -y ppa:deadsnakes/ppa && break || \
-        { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
-    done \
+    && add-apt-repository ppa:deadsnakes/ppa \
     && apt-get update -y \
     && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv libibverbs-dev \
     && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
@@ -209,14 +174,6 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
     && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
     && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \
     && python3 --version && python3 -m pip --version
-# Install uv for faster pip installs
-RUN --mount=type=cache,target=/root/.cache/uv \
-    python3 -m pip install uv
-
-# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
-# Reference: https://github.com/astral-sh/uv/pull/1694
-ENV UV_HTTP_TIMEOUT=500
-ENV UV_INDEX_STRATEGY="unsafe-best-match"
 
 # Workaround for https://github.com/openai/triton/issues/2507 and
 # https://github.com/pytorch/pytorch/issues/107960 -- hopefully
@@ -228,53 +185,39 @@ RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/
 # we need to install torch and torchvision from the nightly builds first,
 # pytorch will not appear as a vLLM dependency in all of the following steps
 # after this step
-RUN --mount=type=cache,target=/root/.cache/uv \
+RUN --mount=type=cache,target=/root/.cache/pip \
     if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
-        uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319";  \
-        uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 --pre pytorch_triton==3.3.0+gitab727c40; \
+        python3 -m pip install --index-url https://download.pytorch.org/whl/nightly/cu124 "torch==2.6.0.dev20241210+cu124" "torchvision==0.22.0.dev20241215";  \
     fi
 
 # Install vllm wheel first, so that torch etc will be installed.
 RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \
-    --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system dist/*.whl --verbose \
-    --extra-index-url https://download.pytorch.org/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+    --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install dist/*.whl --verbose
 
-# If we need to build FlashInfer wheel before its release:
+# How to build this FlashInfer wheel:
 # $ export FLASHINFER_ENABLE_AOT=1
 # $ # Note we remove 7.0 from the arch list compared to the list below, since FlashInfer only supports sm75+
 # $ export TORCH_CUDA_ARCH_LIST='7.5 8.0 8.6 8.9 9.0+PTX'
 # $ git clone https://github.com/flashinfer-ai/flashinfer.git --recursive
 # $ cd flashinfer
 # $ git checkout 524304395bd1d8cd7d07db083859523fcaa246a4
-# $ rm -rf build
 # $ python3 setup.py bdist_wheel --dist-dir=dist --verbose
-# $ ls dist
-# $ # upload the wheel to a public location, e.g. https://wheels.vllm.ai/flashinfer/524304395bd1d8cd7d07db083859523fcaa246a4/flashinfer_python-0.2.1.post1+cu124torch2.5-cp38-abi3-linux_x86_64.whl
 
-RUN --mount=type=cache,target=/root/.cache/uv \
+RUN --mount=type=cache,target=/root/.cache/pip \
 . /etc/environment && \
 if [ "$TARGETPLATFORM" != "linux/arm64" ]; then \
-    # TESTING: install FlashInfer from source to test 2.7.0 final RC
-    FLASHINFER_ENABLE_AOT=1 TORCH_CUDA_ARCH_LIST='7.5 8.0 8.6 8.9 9.0+PTX' \
-    uv pip install --system --no-build-isolation "git+https://github.com/flashinfer-ai/flashinfer@v0.2.2.post1" ; \
+    python3 -m pip install https://wheels.vllm.ai/flashinfer/524304395bd1d8cd7d07db083859523fcaa246a4/flashinfer_python-0.2.0.post1-cp${PYTHON_VERSION_STR}-cp${PYTHON_VERSION_STR}-linux_x86_64.whl; \
 fi
 COPY examples examples
-COPY benchmarks benchmarks
-COPY ./vllm/collect_env.py .
-
-RUN --mount=type=cache,target=/root/.cache/uv \
-. /etc/environment && \
-uv pip list
 
 # Although we build Flashinfer with AOT mode, there's still
 # some issues w.r.t. JIT compilation. Therefore we need to
 # install build dependencies for JIT compilation.
 # TODO: Remove this once FlashInfer AOT wheel is fixed
-COPY requirements/build.txt requirements/build.txt
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system -r requirements/build.txt \
-    --extra-index-url https://download.pytorch.org/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+COPY requirements-build.txt requirements-build.txt
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install -r requirements-build.txt
 
 #################### vLLM installation IMAGE ####################
 
@@ -285,26 +228,17 @@ FROM vllm-base AS test
 
 ADD . /vllm-workspace/
 
-# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
-# Reference: https://github.com/astral-sh/uv/pull/1694
-ENV UV_HTTP_TIMEOUT=500
-ENV UV_INDEX_STRATEGY="unsafe-best-match"
-
-# Workaround for #17068
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system --no-build-isolation "git+https://github.com/state-spaces/mamba@v2.2.4"
+# install development dependencies (for testing)
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install -r requirements-dev.txt
 
 # install development dependencies (for testing)
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system -r requirements/dev.txt
-
-# install development dependencies (for testing)
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system -e tests/vllm_test_utils
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install -e tests/vllm_test_utils
 
 # enable fast downloads from hf (for testing)
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system hf_transfer
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m pip install hf_transfer
 ENV HF_HUB_ENABLE_HF_TRANSFER 1
 
 # Copy in the v1 package for testing (it isn't distributed yet)
@@ -321,18 +255,13 @@ RUN mv vllm test_docs/
 #################### OPENAI API SERVER ####################
 # base openai image with additional requirements, for any subsequent openai-style images
 FROM vllm-base AS vllm-openai-base
-ARG TARGETPLATFORM
-
-# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
-# Reference: https://github.com/astral-sh/uv/pull/1694
-ENV UV_HTTP_TIMEOUT=500
 
 # install additional dependencies for openai api server
-RUN --mount=type=cache,target=/root/.cache/uv \
+RUN --mount=type=cache,target=/root/.cache/pip \
     if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
-        uv pip install --system accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.42.0' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \
+        pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.42.0' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \
     else \
-        uv pip install --system accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.45.3' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \
+        pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.45.0' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \
     fi
 
 ENV VLLM_USAGE_SOURCE production-docker-image

Dockerfile.arm

@@ -26,18 +26,18 @@ WORKDIR /workspace
 ARG PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu"
 ENV PIP_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}
 RUN --mount=type=cache,target=/root/.cache/pip \
-    --mount=type=bind,src=requirements/build.txt,target=requirements/build.txt \
+    --mount=type=bind,src=requirements-build.txt,target=requirements-build.txt \
     pip install --upgrade pip && \
-    pip install -r requirements/build.txt
+    pip install -r requirements-build.txt
 
 FROM cpu-test-arm AS build
 
 WORKDIR /workspace/vllm
 
 RUN --mount=type=cache,target=/root/.cache/pip \
-    --mount=type=bind,src=requirements/common.txt,target=requirements/common.txt \
-    --mount=type=bind,src=requirements/cpu.txt,target=requirements/cpu.txt \
-    pip install -v -r requirements/cpu.txt
+    --mount=type=bind,src=requirements-common.txt,target=requirements-common.txt \
+    --mount=type=bind,src=requirements-cpu.txt,target=requirements-cpu.txt \
+    pip install -v -r requirements-cpu.txt
 
 COPY . .
 ARG GIT_REPO_CHECK=0

Dockerfile.cpu

@@ -0,0 +1,69 @@
+# This vLLM Dockerfile is used to construct image that can build and run vLLM on x86 CPU platform.
+
+FROM ubuntu:22.04 AS cpu-test-1
+
+ENV CCACHE_DIR=/root/.cache/ccache
+
+ENV CMAKE_CXX_COMPILER_LAUNCHER=ccache
+
+RUN --mount=type=cache,target=/var/cache/apt \
+    apt-get update -y \
+    && apt-get install -y curl ccache git wget vim numactl gcc-12 g++-12 python3 python3-pip libtcmalloc-minimal4 libnuma-dev \
+    && apt-get install -y ffmpeg libsm6 libxext6 libgl1 \
+    && update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 10 --slave /usr/bin/g++ g++ /usr/bin/g++-12
+
+# https://intel.github.io/intel-extension-for-pytorch/cpu/latest/tutorials/performance_tuning/tuning_guide.html
+# intel-openmp provides additional performance improvement vs. openmp
+# tcmalloc provides better memory allocation efficiency, e.g, holding memory in caches to speed up access of commonly-used objects.
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install intel-openmp==2025.0.1
+
+ENV LD_PRELOAD="/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4:/usr/local/lib/libiomp5.so"
+
+RUN echo 'ulimit -c 0' >> ~/.bashrc
+
+RUN pip install intel_extension_for_pytorch==2.5.0
+
+WORKDIR /workspace
+
+ARG PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu"
+ENV PIP_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}
+RUN --mount=type=cache,target=/root/.cache/pip \
+    --mount=type=bind,src=requirements-build.txt,target=requirements-build.txt \
+    pip install --upgrade pip && \
+    pip install -r requirements-build.txt
+
+FROM cpu-test-1 AS build
+
+WORKDIR /workspace/vllm
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    --mount=type=bind,src=requirements-common.txt,target=requirements-common.txt \
+    --mount=type=bind,src=requirements-cpu.txt,target=requirements-cpu.txt \
+    pip install -v -r requirements-cpu.txt
+
+COPY . .
+ARG GIT_REPO_CHECK=0
+RUN --mount=type=bind,source=.git,target=.git \
+    if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh ; fi
+
+# Support for building with non-AVX512 vLLM: docker build --build-arg VLLM_CPU_DISABLE_AVX512="true" ...
+ARG VLLM_CPU_DISABLE_AVX512
+ENV VLLM_CPU_DISABLE_AVX512=${VLLM_CPU_DISABLE_AVX512}
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    --mount=type=cache,target=/root/.cache/ccache \
+    --mount=type=bind,source=.git,target=.git \
+    VLLM_TARGET_DEVICE=cpu python3 setup.py bdist_wheel && \
+    pip install dist/*.whl && \
+    rm -rf dist
+
+WORKDIR /workspace/
+
+RUN ln -s /workspace/vllm/tests && ln -s /workspace/vllm/examples && ln -s /workspace/vllm/benchmarks
+
+# install development dependencies (for testing)
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install -e tests/vllm_test_utils
+
+ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]

Dockerfile.hpu

@@ -1,10 +1,10 @@
-FROM vault.habana.ai/gaudi-docker/1.20.1/ubuntu22.04/habanalabs/pytorch-installer-2.6.0:latest
+FROM vault.habana.ai/gaudi-docker/1.19.1/ubuntu22.04/habanalabs/pytorch-installer-2.5.1:latest
 
 COPY ./ /workspace/vllm
 
 WORKDIR /workspace/vllm
 
-RUN pip install -v -r requirements/hpu.txt
+RUN pip install -v -r requirements-hpu.txt
 
 ENV no_proxy=localhost,127.0.0.1
 ENV PT_HPU_ENABLE_LAZY_COLLECTIVES=true

Dockerfile.neuron

@@ -1,6 +1,6 @@
 # default base image
 # https://gallery.ecr.aws/neuron/pytorch-inference-neuronx
-ARG BASE_IMAGE="public.ecr.aws/neuron/pytorch-inference-neuronx:2.5.1-neuronx-py310-sdk2.22.0-ubuntu22.04"
+ARG BASE_IMAGE="public.ecr.aws/neuron/pytorch-inference-neuronx:2.5.1-neuronx-py310-sdk2.21.0-ubuntu22.04"
 
 FROM $BASE_IMAGE
 
@@ -21,9 +21,9 @@ VOLUME [ ${APP_MOUNT} ]
 WORKDIR ${APP_MOUNT}/vllm
 
 RUN python3 -m pip install --upgrade pip
-RUN python3 -m pip install --no-cache-dir fastapi ninja tokenizers pandas tenacity
-RUN python3 -m pip install sentencepiece transformers==4.48.0 -U
-RUN python3 -m pip install neuronx-cc==2.17.194.0 --extra-index-url=https://pip.repos.neuron.amazonaws.com -U
+RUN python3 -m pip install --no-cache-dir fastapi ninja tokenizers pandas
+RUN python3 -m pip install sentencepiece transformers==4.45.2 -U
+RUN python3 -m pip install neuronx-cc==2.16.345.0 --extra-index-url=https://pip.repos.neuron.amazonaws.com -U
 RUN python3 -m pip install pytest
 
 # uninstall transformers-neuronx package explicitly to avoid version conflict
@@ -36,7 +36,7 @@ RUN --mount=type=bind,source=.git,target=.git \
 
 RUN python3 -m pip install -U \
         'cmake>=3.26' ninja packaging 'setuptools-scm>=8' wheel jinja2 \
-        -r requirements/neuron.txt
+        -r requirements-neuron.txt
 
 ENV VLLM_TARGET_DEVICE neuron
 RUN --mount=type=bind,source=.git,target=.git \

Dockerfile.openvino

@@ -0,0 +1,29 @@
+# The vLLM Dockerfile is used to construct vLLM image that can be directly used
+# to run the OpenAI compatible server.
+
+FROM ubuntu:22.04 AS dev
+
+RUN apt-get update -y && \
+    apt-get install -y \
+        git python3-pip \
+        ffmpeg libsm6 libxext6 libgl1
+WORKDIR /workspace
+
+COPY . .
+ARG GIT_REPO_CHECK=0
+RUN --mount=type=bind,source=.git,target=.git \
+    if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh ; fi
+
+RUN python3 -m pip install -U pip
+# install build requirements
+RUN PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" python3 -m pip install -r /workspace/requirements-build.txt
+# build vLLM with OpenVINO backend
+RUN PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" VLLM_TARGET_DEVICE="openvino" python3 -m pip install /workspace
+
+COPY examples/ /workspace/examples
+COPY benchmarks/ /workspace/benchmarks
+
+# install development dependencies (for testing)
+RUN python3 -m pip install -e tests/vllm_test_utils
+
+CMD ["/bin/bash"]

Dockerfile.ppc64le

@@ -0,0 +1,37 @@
+FROM mambaorg/micromamba
+ARG MAMBA_DOCKERFILE_ACTIVATE=1
+USER root
+
+ENV PATH="/usr/local/cargo/bin:$PATH:/opt/conda/bin/"
+
+RUN apt-get update -y && apt-get install -y git wget kmod curl vim libnuma-dev libsndfile-dev libprotobuf-dev build-essential ffmpeg libsm6 libxext6 libgl1 libssl-dev 
+
+# Some packages in requirements-cpu are installed here
+# IBM provides optimized packages for ppc64le processors in the open-ce project for mamba
+# Currently these may not be available for venv or pip directly
+RUN micromamba install -y -n base -c https://ftp.osuosl.org/pub/open-ce/1.11.0-p10/ -c defaults python=3.10 rust && micromamba clean --all --yes
+
+COPY ./ /workspace/vllm
+
+WORKDIR /workspace/vllm
+ARG GIT_REPO_CHECK=0
+RUN --mount=type=bind,source=.git,target=.git \
+    if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
+
+RUN --mount=type=cache,target=/root/.cache/pip  \
+    RUSTFLAGS='-L /opt/conda/lib' pip install -v --prefer-binary --extra-index-url https://repo.fury.io/mgiessing \
+        'cmake>=3.26' ninja packaging 'setuptools-scm>=8' wheel jinja2 \
+        -r requirements-cpu.txt \
+        xformers uvloop==0.20.0
+
+RUN --mount=type=bind,source=.git,target=.git \
+    VLLM_TARGET_DEVICE=cpu python3 setup.py install
+
+# install development dependencies (for testing)
+RUN python3 -m pip install -e tests/vllm_test_utils
+
+WORKDIR /workspace/
+
+RUN ln -s /workspace/vllm/tests && ln -s /workspace/vllm/examples && ln -s /workspace/vllm/benchmarks
+
+ENTRYPOINT ["/opt/conda/bin/python3", "-m", "vllm.entrypoints.openai.api_server"]

Dockerfile.rocm

@@ -12,8 +12,7 @@ ENV PYTORCH_ROCM_ARCH=${ARG_PYTORCH_ROCM_ARCH:-${PYTORCH_ROCM_ARCH}}
 
 # Install some basic utilities
 RUN apt-get update -q -y && apt-get install -q -y \
-    sqlite3 libsqlite3-dev libfmt-dev libmsgpack-dev libsuitesparse-dev \
-    apt-transport-https ca-certificates wget curl
+    sqlite3 libsqlite3-dev libfmt-dev libmsgpack-dev libsuitesparse-dev
 # Remove sccache    
 RUN python3 -m pip install --upgrade pip && pip install setuptools_scm
 RUN apt-get purge -y sccache; python3 -m pip uninstall -y sccache; rm -f "$(which sccache)"
@@ -39,14 +38,14 @@ FROM fetch_vllm AS build_vllm
 ARG USE_CYTHON
 # Build vLLM
 RUN cd vllm \
-    && python3 -m pip install -r requirements/rocm.txt \
+    && python3 -m pip install -r requirements-rocm.txt \
     && python3 setup.py clean --all  \
-    && if [ ${USE_CYTHON} -eq "1" ]; then python3 tests/build_cython.py build_ext --inplace; fi \
+    && if [ ${USE_CYTHON} -eq "1" ]; then python3 setup_cython.py build_ext --inplace; fi \
     && python3 setup.py bdist_wheel --dist-dir=dist
 FROM scratch AS export_vllm
 ARG COMMON_WORKDIR
 COPY --from=build_vllm ${COMMON_WORKDIR}/vllm/dist/*.whl /
-COPY --from=build_vllm ${COMMON_WORKDIR}/vllm/requirements /requirements
+COPY --from=build_vllm ${COMMON_WORKDIR}/vllm/requirements*.txt /
 COPY --from=build_vllm ${COMMON_WORKDIR}/vllm/benchmarks /benchmarks
 COPY --from=build_vllm ${COMMON_WORKDIR}/vllm/tests /tests
 COPY --from=build_vllm ${COMMON_WORKDIR}/vllm/examples /examples
@@ -61,8 +60,7 @@ RUN python3 -m pip install --upgrade pip && rm -rf /var/lib/apt/lists/*
 # Install vLLM
 RUN --mount=type=bind,from=export_vllm,src=/,target=/install \
     cd /install \
-    && pip install -U -r requirements/rocm.txt \
-    && pip install -U -r requirements/rocm-test.txt \
+    && pip install -U -r requirements-rocm.txt \
     && pip uninstall -y vllm \
     && pip install *.whl
 
@@ -101,7 +99,7 @@ RUN if [ ${BUILD_RPD} -eq "1" ]; then \
 # Install vLLM
 RUN --mount=type=bind,from=export_vllm,src=/,target=/install \
     cd /install \
-    && pip install -U -r requirements/rocm.txt \
+    && pip install -U -r requirements-rocm.txt \
     && pip uninstall -y vllm \
     && pip install *.whl
 
@@ -114,16 +112,8 @@ COPY --from=export_vllm /examples ${COMMON_WORKDIR}/vllm/examples
 ENV RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
 ENV TOKENIZERS_PARALLELISM=false
 
-# ENV that can improve safe tensor loading, and end-to-end time
-ENV SAFETENSORS_FAST_GPU=1
-
-# User-friendly environment setting for multi-processing to avoid below RuntimeError.
-# RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing,
-# you must use the 'spawn' start method 
-# See https://pytorch.org/docs/stable/notes/multiprocessing.html#cuda-in-multiprocessing
-ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
-
 # Performance environment variable.
 ENV HIP_FORCE_DEV_KERNARG=1
 
 CMD ["/bin/bash"]
+

Dockerfile.rocm_base

@@ -1,26 +1,24 @@
 ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:6.3.1-complete
-ARG HIPBLASLT_BRANCH="db8e93b4"
+ARG HIPBLASLT_BRANCH="4d40e36"
 ARG HIPBLAS_COMMON_BRANCH="7c1566b"
 ARG LEGACY_HIPBLASLT_OPTION=
 ARG RCCL_BRANCH="648a58d"
 ARG RCCL_REPO="https://github.com/ROCm/rccl"
 ARG TRITON_BRANCH="e5be006"
 ARG TRITON_REPO="https://github.com/triton-lang/triton.git"
-ARG PYTORCH_BRANCH="295f2ed4"
-ARG PYTORCH_VISION_BRANCH="v0.21.0"
+ARG PYTORCH_BRANCH="3a585126"
+ARG PYTORCH_VISION_BRANCH="v0.19.1"
 ARG PYTORCH_REPO="https://github.com/pytorch/pytorch.git"
 ARG PYTORCH_VISION_REPO="https://github.com/pytorch/vision.git"
-ARG FA_BRANCH="1a7f4dfa"
-ARG FA_REPO="https://github.com/Dao-AILab/flash-attention.git"
-ARG AITER_BRANCH="5a77249"
-ARG AITER_REPO="https://github.com/ROCm/aiter.git"
+ARG FA_BRANCH="b7d29fb"
+ARG FA_REPO="https://github.com/ROCm/flash-attention.git"
 
 FROM ${BASE_IMAGE} AS base
 
 ENV PATH=/opt/rocm/llvm/bin:$PATH
 ENV ROCM_PATH=/opt/rocm
 ENV LD_LIBRARY_PATH=/opt/rocm/lib:/usr/local/lib:
-ARG PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx1100;gfx1101;gfx1200;gfx1201
+ARG PYTORCH_ROCM_ARCH=gfx90a;gfx942
 ENV PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH}
 
 ARG PYTHON_VERSION=3.12
@@ -31,11 +29,8 @@ ENV DEBIAN_FRONTEND=noninteractive
 
 # Install Python and other dependencies
 RUN apt-get update -y \
-    && apt-get install -y software-properties-common git curl sudo vim less libgfortran5 \
-    && for i in 1 2 3; do \
-        add-apt-repository -y ppa:deadsnakes/ppa && break || \
-        { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
-    done \
+    && apt-get install -y software-properties-common git curl sudo vim less \
+    && add-apt-repository ppa:deadsnakes/ppa \
     && apt-get update -y \
     && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
        python${PYTHON_VERSION}-lib2to3 python-is-python3  \
@@ -45,7 +40,7 @@ RUN apt-get update -y \
     && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \
     && python3 --version && python3 -m pip --version
 
-RUN pip install -U packaging 'cmake<4' ninja wheel setuptools pybind11 Cython
+RUN pip install -U packaging cmake ninja wheel setuptools pybind11 Cython
 
 FROM base AS build_hipblaslt
 ARG HIPBLASLT_BRANCH
@@ -63,8 +58,7 @@ RUN cd hipBLAS-common \
 RUN git clone https://github.com/ROCm/hipBLASLt
 RUN cd hipBLASLt \
     && git checkout ${HIPBLASLT_BRANCH} \
-    && apt-get install -y llvm-dev \
-    && ./install.sh -dc --architecture ${PYTORCH_ROCM_ARCH} ${LEGACY_HIPBLASLT_OPTION} \
+    && ./install.sh -d --architecture ${PYTORCH_ROCM_ARCH} ${LEGACY_HIPBLASLT_OPTION} \
     && cd build/release \
     && make package
 RUN mkdir -p /app/install && cp /app/hipBLASLt/build/release/*.deb /app/hipBLAS-common/build/*.deb /app/install
@@ -114,24 +108,11 @@ RUN git clone ${FA_REPO}
 RUN cd flash-attention \
     && git checkout ${FA_BRANCH} \
     && git submodule update --init \
-    && GPU_ARCHS=$(echo ${PYTORCH_ROCM_ARCH} | sed -e 's/;gfx1[0-9]\{3\}//g') python3 setup.py bdist_wheel --dist-dir=dist
+    && MAX_JOBS=64 GPU_ARCHS=${PYTORCH_ROCM_ARCH} python3 setup.py bdist_wheel --dist-dir=dist
 RUN mkdir -p /app/install && cp /app/pytorch/dist/*.whl /app/install \
     && cp /app/vision/dist/*.whl /app/install \
     && cp /app/flash-attention/dist/*.whl /app/install
 
-FROM base AS build_aiter
-ARG AITER_BRANCH
-ARG AITER_REPO
-RUN --mount=type=bind,from=build_pytorch,src=/app/install/,target=/install \
-    pip install /install/*.whl
-RUN git clone --recursive ${AITER_REPO}
-RUN cd aiter \
-    && git checkout ${AITER_BRANCH} \
-    && git submodule update --init --recursive \
-    && pip install -r requirements.txt
-RUN pip install pyyaml && cd aiter && PREBUILD_KERNELS=1 GPU_ARCHS=gfx942 python3 setup.py bdist_wheel --dist-dir=dist && ls /app/aiter/dist/*.whl
-RUN mkdir -p /app/install && cp /app/aiter/dist/*.whl /app/install
-
 FROM base AS final
 RUN --mount=type=bind,from=build_hipblaslt,src=/app/install/,target=/install \
     dpkg -i /install/*deb \
@@ -147,11 +128,8 @@ RUN --mount=type=bind,from=build_amdsmi,src=/app/install/,target=/install \
     pip install /install/*.whl
 RUN --mount=type=bind,from=build_pytorch,src=/app/install/,target=/install \
     pip install /install/*.whl
-RUN --mount=type=bind,from=build_aiter,src=/app/install/,target=/install \
-    pip install /install/*.whl
 
 ARG BASE_IMAGE
-ARG HIPBLAS_COMMON_BRANCH
 ARG HIPBLASLT_BRANCH
 ARG LEGACY_HIPBLASLT_OPTION
 ARG RCCL_BRANCH
@@ -164,8 +142,6 @@ ARG PYTORCH_REPO
 ARG PYTORCH_VISION_REPO
 ARG FA_BRANCH
 ARG FA_REPO
-ARG AITER_BRANCH
-ARG AITER_REPO
 RUN echo "BASE_IMAGE: ${BASE_IMAGE}" > /app/versions.txt \
     && echo "HIPBLAS_COMMON_BRANCH: ${HIPBLAS_COMMON_BRANCH}" >> /app/versions.txt \
     && echo "HIPBLASLT_BRANCH: ${HIPBLASLT_BRANCH}" >> /app/versions.txt \
@@ -179,5 +155,4 @@ RUN echo "BASE_IMAGE: ${BASE_IMAGE}" > /app/versions.txt \
     && echo "PYTORCH_REPO: ${PYTORCH_REPO}" >> /app/versions.txt \
     && echo "PYTORCH_VISION_REPO: ${PYTORCH_VISION_REPO}" >> /app/versions.txt \
     && echo "FA_BRANCH: ${FA_BRANCH}" >> /app/versions.txt \
-    && echo "AITER_BRANCH: ${AITER_BRANCH}" >> /app/versions.txt \
-    && echo "AITER_REPO: ${AITER_REPO}" >> /app/versions.txt
+    && echo "FA_REPO: ${FA_REPO}" >> /app/versions.txt

Dockerfile.tpu

@@ -15,15 +15,12 @@ ARG GIT_REPO_CHECK=0
 RUN --mount=type=bind,source=.git,target=.git \
     if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
 
-# Remove existing versions of dependencies
-RUN pip uninstall -y torch torch_xla torchvision
-
 ENV VLLM_TARGET_DEVICE="tpu"
 RUN --mount=type=cache,target=/root/.cache/pip \
     --mount=type=bind,source=.git,target=.git \
     python3 -m pip install \
-        -r requirements/tpu.txt
-RUN python3 -m pip install -e .
+        -r requirements-tpu.txt
+RUN python3 setup.py develop
 
 # install development dependencies (for testing)
 RUN python3 -m pip install -e tests/vllm_test_utils

Dockerfile.xpu

@@ -0,0 +1,69 @@
+FROM intel/oneapi-basekit:2024.2.1-0-devel-ubuntu22.04 AS vllm-base
+
+RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | tee /usr/share/keyrings/intel-oneapi-archive-keyring.gpg > /dev/null && \
+    echo "deb [signed-by=/usr/share/keyrings/intel-oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main " | tee /etc/apt/sources.list.d/oneAPI.list && \
+    chmod 644 /usr/share/keyrings/intel-oneapi-archive-keyring.gpg && \
+    wget -O- https://repositories.intel.com/graphics/intel-graphics.key | gpg --dearmor | tee /usr/share/keyrings/intel-graphics.gpg > /dev/null && \
+    echo "deb [arch=amd64,i386 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/graphics/ubuntu jammy arc" | tee /etc/apt/sources.list.d/intel.gpu.jammy.list && \
+    chmod 644 /usr/share/keyrings/intel-graphics.gpg
+
+RUN apt-get update -y && \
+    apt-get install -y --no-install-recommends --fix-missing \
+    curl \
+    ffmpeg \
+    git \
+    libsndfile1 \
+    libsm6 \
+    libxext6 \
+    libgl1 \
+    lsb-release \
+    numactl \
+    python3 \
+    python3-dev \
+    python3-pip \
+    # vim \
+    wget
+
+WORKDIR /workspace/vllm
+COPY requirements-xpu.txt /workspace/vllm/requirements-xpu.txt
+COPY requirements-common.txt /workspace/vllm/requirements-common.txt
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install --no-cache-dir \
+    -r requirements-xpu.txt
+
+RUN git clone https://github.com/intel/pti-gpu && \
+    cd pti-gpu/sdk && \
+    git checkout 6c491f07a777ed872c2654ca9942f1d0dde0a082 && \
+    mkdir build && \
+    cd build && \
+    cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_TOOLCHAIN_FILE=../cmake/toolchains/icpx_toolchain.cmake -DBUILD_TESTING=OFF .. && \
+    make -j && \
+    cmake --install . --config Release --prefix "/usr/local"
+
+ENV LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/lib/"
+
+COPY . .
+ARG GIT_REPO_CHECK
+RUN --mount=type=bind,source=.git,target=.git \
+    if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
+
+ENV VLLM_TARGET_DEVICE=xpu
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    --mount=type=bind,source=.git,target=.git \
+    python3 setup.py install
+
+CMD ["/bin/bash"]
+
+FROM vllm-base AS vllm-openai
+
+# install additional dependencies for openai api server
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install accelerate hf_transfer 'modelscope!=1.15.0'
+
+ENV VLLM_USAGE_SOURCE production-docker-image \
+    TRITON_XPU_PROFILE 1
+# install development dependencies (for testing)
+RUN python3 -m pip install -e tests/vllm_test_utils
+ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]

MANIFEST.in

@@ -1,9 +1,9 @@
 include LICENSE
-include requirements/common.txt
-include requirements/cuda.txt
-include requirements/rocm.txt
-include requirements/neuron.txt
-include requirements/cpu.txt
+include requirements-common.txt
+include requirements-cuda.txt
+include requirements-rocm.txt
+include requirements-neuron.txt
+include requirements-cpu.txt
 include CMakeLists.txt
 
 recursive-include cmake *

README.md

@@ -10,24 +10,18 @@ Easy, fast, and cheap LLM serving for everyone
 </h3>
 
 <p align="center">
-| <a href="https://docs.vllm.ai"><b>Documentation</b></a> | <a href="https://blog.vllm.ai/"><b>Blog</b></a> | <a href="https://arxiv.org/abs/2309.06180"><b>Paper</b></a> | <a href="https://x.com/vllm_project"><b>Twitter/X</b></a> | <a href="https://discuss.vllm.ai"><b>User Forum</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> |
+| <a href="https://docs.vllm.ai"><b>Documentation</b></a> | <a href="https://vllm.ai"><b>Blog</b></a> | <a href="https://arxiv.org/abs/2309.06180"><b>Paper</b></a> | <a href="https://x.com/vllm_project"><b>Twitter/X</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> |
 </p>
 
 ---
 
+We are excited to invite you to our Menlo Park meetup with Meta, evening of Thursday, February 27! Meta engineers will discuss the improvements on top of vLLM, and vLLM contributors will share updates from the v0.7.x series of releases. [Register Now](https://lu.ma/h7g3kuj9)
+
+---
+
 *Latest News* 🔥
-- [2025/05] We hosted [NYC vLLM Meetup](https://lu.ma/c1rqyf1f)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1_q_aW_ioMJWUImf1s1YM-ZhjXz8cUeL0IJvaquOYBeA/edit?usp=sharing).
-- [2025/05] vLLM is now a hosted project under PyTorch Foundation! Please find the announcement [here](https://pytorch.org/blog/pytorch-foundation-welcomes-vllm/).
-- [2025/04] We hosted [Asia Developer Day](https://www.sginnovate.com/event/limited-availability-morning-evening-slots-remaining-inaugural-vllm-asia-developer-day)! Please find the meetup slides from the vLLM team [here](https://docs.google.com/presentation/d/19cp6Qu8u48ihB91A064XfaXruNYiBOUKrBxAmDOllOo/edit?usp=sharing).
+
 - [2025/01] We are excited to announce the alpha release of vLLM V1: A major architectural upgrade with 1.7x speedup! Clean code, optimized execution loop, zero-overhead prefix caching, enhanced multimodal support, and more. Please check out our blog post [here](https://blog.vllm.ai/2025/01/27/v1-alpha-release.html).
-
-<details>
-<summary>Previous News</summary>
-
-- [2025/03] We hosted [vLLM x Ollama Inference Night](https://lu.ma/vllm-ollama)! Please find the meetup slides from the vLLM team [here](https://docs.google.com/presentation/d/16T2PDD1YwRnZ4Tu8Q5r6n53c5Lr5c73UV9Vd2_eBo4U/edit?usp=sharing).
-- [2025/03] We hosted [the first vLLM China Meetup](https://mp.weixin.qq.com/s/n77GibL2corAtQHtVEAzfg)! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1REHvfQMKGnvz6p3Fd23HhSO4c8j5WPGZV0bKYLwnHyQ/edit?usp=sharing).
-- [2025/03] We hosted [the East Coast vLLM Meetup](https://lu.ma/7mu4k4xx)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1NHiv8EUFF1NLd3fEYODm56nDmL26lEeXCaDgyDlTsRs/edit#slide=id.g31441846c39_0_0).
-- [2025/02] We hosted [the ninth vLLM meetup](https://lu.ma/h7g3kuj9) with Meta! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1jzC_PZVXrVNSFVCW-V4cFXb6pn7zZ2CyP_Flwo05aqg/edit?usp=sharing) and AMD [here](https://drive.google.com/file/d/1Zk5qEJIkTmlQ2eQcXQZlljAx3m9s7nwn/view?usp=sharing). The slides from Meta will not be posted.
 - [2025/01] We hosted [the eighth vLLM meetup](https://lu.ma/zep56hui) with Google Cloud! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1epVkt4Zu8Jz_S5OhEHPc798emsYh2BwYfRuDDVEF7u4/edit?usp=sharing), and Google Cloud team [here](https://drive.google.com/file/d/1h24pHewANyRL11xy5dXUbvRC9F9Kkjix/view?usp=sharing).
 - [2024/12] vLLM joins [pytorch ecosystem](https://pytorch.org/blog/vllm-joins-pytorch)! Easy, Fast, and Cheap LLM Serving for Everyone!
 - [2024/11] We hosted [the seventh vLLM meetup](https://lu.ma/h0qvrajz) with Snowflake! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1e3CxQBV3JsfGp30SwyvS3eM_tW-ghOhJ9PAJGK6KR54/edit?usp=sharing), and Snowflake team [here](https://docs.google.com/presentation/d/1qF3RkDAbOULwz9WK5TOltt2fE9t6uIc_hVNLFAaQX6A/edit?usp=sharing).
@@ -43,9 +37,8 @@ Easy, fast, and cheap LLM serving for everyone
 - [2023/08] We would like to express our sincere gratitude to [Andreessen Horowitz](https://a16z.com/2023/08/30/supporting-the-open-source-ai-community/) (a16z) for providing a generous grant to support the open-source development and research of vLLM.
 - [2023/06] We officially released vLLM! FastChat-vLLM integration has powered [LMSYS Vicuna and Chatbot Arena](https://chat.lmsys.org) since mid-April. Check out our [blog post](https://vllm.ai).
 
-</details>
-
 ---
+
 ## About
 
 vLLM is a fast and easy-to-use library for LLM inference and serving.
@@ -93,14 +86,14 @@ pip install vllm
 ```
 
 Visit our [documentation](https://docs.vllm.ai/en/latest/) to learn more.
-- [Installation](https://docs.vllm.ai/en/latest/getting_started/installation.html)
+- [Installation](https://docs.vllm.ai/en/latest/getting_started/installation/index.html)
 - [Quickstart](https://docs.vllm.ai/en/latest/getting_started/quickstart.html)
 - [List of Supported Models](https://docs.vllm.ai/en/latest/models/supported_models.html)
 
 ## Contributing
 
 We welcome and value any contributions and collaborations.
-Please check out [Contributing to vLLM](https://docs.vllm.ai/en/stable/contributing/overview.html) for how to get involved.
+Please check out [CONTRIBUTING.md](./CONTRIBUTING.md) for how to get involved.
 
 ## Sponsors
 
@@ -123,7 +116,6 @@ Compute Resources:
 - Databricks
 - DeepInfra
 - Google Cloud
-- Intel
 - Lambda Lab
 - Nebius
 - Novita AI
@@ -154,11 +146,10 @@ If you use vLLM for your research, please cite our [paper](https://arxiv.org/abs
 
 ## Contact Us
 
-- For technical questions and feature requests, please use GitHub [Issues](https://github.com/vllm-project/vllm/issues) or [Discussions](https://github.com/vllm-project/vllm/discussions)
-- For discussing with fellow users, please use the [vLLM Forum](https://discuss.vllm.ai)
-- coordinating contributions and development, please use [Slack](https://slack.vllm.ai)
-- For security disclosures, please use GitHub's [Security Advisories](https://github.com/vllm-project/vllm/security/advisories) feature
-- For collaborations and partnerships, please contact us at [vllm-questions@lists.berkeley.edu](mailto:vllm-questions@lists.berkeley.edu)
+- For technical questions and feature requests, please use Github issues or discussions.
+- For discussing with fellow users and coordinating contributions and development, please use Slack.
+- For security disclosures, please use Github's security advisory feature.
+- For collaborations and partnerships, please contact us at vllm-questions AT lists.berkeley.edu.
 
 ## Media Kit
 

RELEASE.md

@@ -1,54 +0,0 @@
-# Releasing vLLM
-
-vLLM releases offer a reliable version of the code base, packaged into a binary format that can be conveniently accessed via PyPI. These releases also serve as key milestones for the development team to communicate with the community about newly available features, improvements, and upcoming changes that could affect users, including potential breaking changes.
-
-## Release Versioning
-
-vLLM uses a “right-shifted” versioning scheme where a new patch release is out every 2 weeks. And patch releases contain features and bug fixes (as opposed to semver where patch release contains only backwards-compatible bug fixes). When critical fixes need to be made, special release post1 is released.
-
-* _major_ major architectural milestone and when incompatible API changes are made, similar to PyTorch 2.0.
-* _minor_ major features
-* _patch_ features and backwards-compatible bug fixes
-* _post1_ or _patch-1_ backwards-compatible bug fixes, either explicit or implicit post release
-
-## Release Cadence
-
-Patch release is released on bi-weekly basis. Post release 1-3 days after patch release and uses same branch as patch release.
-Following is the release cadence for year 2025. All future release dates below are tentative. Please note: Post releases are optional.
-
-| Release Date | Patch release versions | Post Release versions |
-| --- | --- | --- |
-| Jan 2025 | 0.7.0 | --- |
-| Feb 2025 | 0.7.1, 0.7.2, 0.7.3  | --- |
-| Mar 2025 | 0.7.4, 0.7.5 | --- |
-| Apr 2025 | 0.7.6, 0.7.7 | --- |
-| May 2025 | 0.7.8, 0.7.9 | --- |
-| Jun 2025 | 0.7.10, 0.7.11 | --- |
-| Jul 2025 | 0.7.12, 0.7.13 | --- |
-| Aug 2025 | 0.7.14, 0.7.15 | --- |
-| Sep 2025 | 0.7.16, 0.7.17 | --- |
-| Oct 2025 | 0.7.18, 0.7.19 | --- |
-| Nov 2025 | 0.7.20, 0.7.21 | --- |
-| Dec 2025 | 0.7.22, 0.7.23 | --- |
-
-## Release branch
-
-Each release is built from a dedicated release branch.
-
-* For _major_, _minor_, _patch_ releases, the release branch cut is performed 1-2 days before release is live.
-* For post releases, previously cut release branch is reused
-* Release builds are triggered via push to RC tag like vX.Y.Z-rc1 . This enables us to build and test multiple RCs for each release.
-* Final tag : vX.Y.Z does not trigger the build but used for Release notes and assets.
-* After branch cut is created we monitor the main branch for any reverts and apply these reverts to a release branch.
-
-## Release Cherry-Pick Criteria
-
-After branch cut, we approach finalizing the release branch with clear criteria on what cherry picks are allowed in. Note: a cherry pick is a process to land a PR in the release branch after branch cut. These are typically limited to ensure that the team has sufficient time to complete a thorough round of testing on a stable code base.
-
-* Regression fixes - that address functional/performance regression against the most recent release (e.g. 0.7.0 for 0.7.1 release)
-* Critical fixes - critical fixes for severe issue such as silent incorrectness, backwards compatibility, crashes, deadlocks, (large) memory leaks
-* Fixes to new features introduced in the most recent release (e.g. 0.7.0 for 0.7.1 release)
-* Documentation improvements
-* Release branch specific changes (e.g. change version identifiers or CI fixes)
-
-Please note: **No feature work allowed for cherry picks**. All PRs that are considered for cherry-picks need to be merged on trunk, the only exception are Release branch specific changes.

benchmarks/README.md

@@ -1,343 +1,29 @@
 # Benchmarking vLLM
 
-This README guides you through running benchmark tests with the extensive
-datasets supported on vLLM. It’s a living document, updated as new features and datasets
-become available.
+## Downloading the ShareGPT dataset
 
-## Dataset Overview
-
-<table style="width:100%; border-collapse: collapse;">
-  <thead>
-    <tr>
-      <th style="width:15%; text-align: left;">Dataset</th>
-      <th style="width:10%; text-align: center;">Online</th>
-      <th style="width:10%; text-align: center;">Offline</th>
-      <th style="width:65%; text-align: left;">Data Path</th>
-    </tr>
-  </thead>
-  <tbody>
-    <tr>
-      <td><strong>ShareGPT</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json</code></td>
-    </tr>
-    <tr>
-      <td><strong>BurstGPT</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>wget https://github.com/HPMLL/BurstGPT/releases/download/v1.1/BurstGPT_without_fails_2.csv</code></td>
-    </tr>
-    <tr>
-      <td><strong>Sonnet</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td>Local file: <code>benchmarks/sonnet.txt</code></td>
-    </tr>
-    <tr>
-      <td><strong>Random</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>synthetic</code></td>
-    </tr>
-    <tr>
-      <td><strong>HuggingFace-VisionArena</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>lmarena-ai/VisionArena-Chat</code></td>
-    </tr>
-    <tr>
-      <td><strong>HuggingFace-InstructCoder</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>likaixin/InstructCoder</code></td>
-    </tr>
-      <tr>
-      <td><strong>HuggingFace-AIMO</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>AI-MO/aimo-validation-aime</code> , <code>AI-MO/NuminaMath-1.5</code>, <code>AI-MO/NuminaMath-CoT</code></td>
-    </tr>
-    <tr>
-      <td><strong>HuggingFace-Other</strong></td>
-      <td style="text-align: center;">✅</td>
-      <td style="text-align: center;">✅</td>
-      <td><code>lmms-lab/LLaVA-OneVision-Data</code>, <code>Aeala/ShareGPT_Vicuna_unfiltered</code></td>
-    </tr>
-  </tbody>
-</table>
-
-✅: supported
-
-🟡: Partial support
-
-🚧: to be supported
-
-**Note**: HuggingFace dataset's `dataset-name` should be set to `hf`
-
----
-## Example - Online Benchmark
-
-First start serving your model
+You can download the dataset by running:
 
 ```bash
-vllm serve NousResearch/Hermes-3-Llama-3.1-8B --disable-log-requests
+wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
 ```
 
-Then run the benchmarking script
+## Downloading the ShareGPT4V dataset
+
+The json file refers to several image datasets (coco, llava, etc.). The benchmark scripts
+will ignore a datapoint if the referred image is missing.
 
 ```bash
-# download dataset
-# wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
-python3 vllm/benchmarks/benchmark_serving.py \
-  --backend vllm \
-  --model NousResearch/Hermes-3-Llama-3.1-8B \
-  --endpoint /v1/completions \
-  --dataset-name sharegpt \
-  --dataset-path <your data path>/ShareGPT_V3_unfiltered_cleaned_split.json \
-  --num-prompts 10
+wget https://huggingface.co/datasets/Lin-Chen/ShareGPT4V/resolve/main/sharegpt4v_instruct_gpt4-vision_cap100k.json
+mkdir coco -p
+wget http://images.cocodataset.org/zips/train2017.zip -O coco/train2017.zip
+unzip coco/train2017.zip -d coco/
 ```
 
-If successful, you will see the following output
+# Downloading the BurstGPT dataset
 
-```
-============ Serving Benchmark Result ============
-Successful requests:                     10        
-Benchmark duration (s):                  5.78      
-Total input tokens:                      1369      
-Total generated tokens:                  2212      
-Request throughput (req/s):              1.73      
-Output token throughput (tok/s):         382.89    
-Total Token throughput (tok/s):          619.85    
----------------Time to First Token----------------
-Mean TTFT (ms):                          71.54     
-Median TTFT (ms):                        73.88     
-P99 TTFT (ms):                           79.49     
------Time per Output Token (excl. 1st token)------
-Mean TPOT (ms):                          7.91      
-Median TPOT (ms):                        7.96      
-P99 TPOT (ms):                           8.03      
----------------Inter-token Latency----------------
-Mean ITL (ms):                           7.74      
-Median ITL (ms):                         7.70      
-P99 ITL (ms):                            8.39      
-==================================================
-```
-
-### VisionArena Benchmark for Vision Language Models
+You can download the BurstGPT v1.1 dataset by running:
 
 ```bash
-# need a model with vision capability here
-vllm serve Qwen/Qwen2-VL-7B-Instruct --disable-log-requests
+wget https://github.com/HPMLL/BurstGPT/releases/download/v1.1/BurstGPT_without_fails_2.csv
 ```
-
-```bash
-python3 vllm/benchmarks/benchmark_serving.py \
-  --backend openai-chat \
-  --model Qwen/Qwen2-VL-7B-Instruct \
-  --endpoint /v1/chat/completions \
-  --dataset-name hf \
-  --dataset-path lmarena-ai/VisionArena-Chat \
-  --hf-split train \
-  --num-prompts 1000
-```
-
-### InstructCoder Benchmark with Speculative Decoding
-
-``` bash
-VLLM_USE_V1=1 vllm serve meta-llama/Meta-Llama-3-8B-Instruct \
-    --speculative-model "[ngram]" \
-    --ngram_prompt_lookup_min 2 \
-    --ngram-prompt-lookup-max 5 \
-    --num_speculative_tokens 5
-```
-
-``` bash
-python3 benchmarks/benchmark_serving.py \
-    --model meta-llama/Meta-Llama-3-8B-Instruct \
-    --dataset-name hf \
-    --dataset-path likaixin/InstructCoder \
-    --num-prompts 2048
-```
-
-### Other HuggingFaceDataset Examples
-
-```bash
-vllm serve Qwen/Qwen2-VL-7B-Instruct --disable-log-requests
-```
-
-**`lmms-lab/LLaVA-OneVision-Data`**
-
-```bash
-python3 vllm/benchmarks/benchmark_serving.py \
-  --backend openai-chat \
-  --model Qwen/Qwen2-VL-7B-Instruct \
-  --endpoint /v1/chat/completions \
-  --dataset-name hf \
-  --dataset-path lmms-lab/LLaVA-OneVision-Data \
-  --hf-split train \
-  --hf-subset "chart2text(cauldron)" \
-  --num-prompts 10
-```
-
-**`Aeala/ShareGPT_Vicuna_unfiltered`**
-
-```bash
-python3 vllm/benchmarks/benchmark_serving.py \
-  --backend openai-chat \
-  --model Qwen/Qwen2-VL-7B-Instruct \
-  --endpoint /v1/chat/completions \
-  --dataset-name hf \
-  --dataset-path Aeala/ShareGPT_Vicuna_unfiltered \
-  --hf-split train \
-  --num-prompts 10
-```
-
-**`AI-MO/aimo-validation-aime`**
-
-``` bash
-python3 vllm/benchmarks/benchmark_serving.py \
-    --model Qwen/QwQ-32B \
-    --dataset-name hf \
-    --dataset-path AI-MO/aimo-validation-aime \
-    --num-prompts 10 \
-    --seed 42
-```
-
-### Running With Sampling Parameters
-
-When using OpenAI-compatible backends such as `vllm`, optional sampling
-parameters can be specified. Example client command:
-
-```bash
-python3 vllm/benchmarks/benchmark_serving.py \
-  --backend vllm \
-  --model NousResearch/Hermes-3-Llama-3.1-8B \
-  --endpoint /v1/completions \
-  --dataset-name sharegpt \
-  --dataset-path <your data path>/ShareGPT_V3_unfiltered_cleaned_split.json \
-  --top-k 10 \
-  --top-p 0.9 \
-  --temperature 0.5 \
-  --num-prompts 10
-```
-
----
-## Example - Offline Throughput Benchmark
-
-```bash
-python3 vllm/benchmarks/benchmark_throughput.py \
-  --model NousResearch/Hermes-3-Llama-3.1-8B \
-  --dataset-name sonnet \
-  --dataset-path vllm/benchmarks/sonnet.txt \
-  --num-prompts 10
-```
-
-If successful, you will see the following output
-
-```
-Throughput: 7.15 requests/s, 4656.00 total tokens/s, 1072.15 output tokens/s
-Total num prompt tokens:  5014
-Total num output tokens:  1500
-```
-
-### VisionArena Benchmark for Vision Language Models
-
-``` bash
-python3 vllm/benchmarks/benchmark_throughput.py \
-  --model Qwen/Qwen2-VL-7B-Instruct \
-  --backend vllm-chat \
-  --dataset-name hf \
-  --dataset-path lmarena-ai/VisionArena-Chat \
-  --num-prompts 1000 \
-  --hf-split train
-```
-
-The `num prompt tokens` now includes image token counts
-
-```
-Throughput: 2.55 requests/s, 4036.92 total tokens/s, 326.90 output tokens/s
-Total num prompt tokens:  14527
-Total num output tokens:  1280
-```
-
-### InstructCoder Benchmark with Speculative Decoding
-
-``` bash
-VLLM_WORKER_MULTIPROC_METHOD=spawn \
-VLLM_USE_V1=1 \
-python3 vllm/benchmarks/benchmark_throughput.py \
-    --dataset-name=hf \
-    --dataset-path=likaixin/InstructCoder \
-    --model=meta-llama/Meta-Llama-3-8B-Instruct \
-    --input-len=1000 \
-    --output-len=100 \
-    --num-prompts=2048 \
-    --async-engine \
-    --speculative-model="[ngram]" \
-    --ngram_prompt_lookup_min=2 \
-    --ngram-prompt-lookup-max=5 \
-    --num_speculative_tokens=5
-```
-
-```
-Throughput: 104.77 requests/s, 23836.22 total tokens/s, 10477.10 output tokens/s
-Total num prompt tokens:  261136
-Total num output tokens:  204800
-```
-
-### Other HuggingFaceDataset Examples
-
-**`lmms-lab/LLaVA-OneVision-Data`**
-
-```bash
-python3 vllm/benchmarks/benchmark_throughput.py \
-  --model Qwen/Qwen2-VL-7B-Instruct \
-  --backend vllm-chat \
-  --dataset-name hf \
-  --dataset-path lmms-lab/LLaVA-OneVision-Data \
-  --hf-split train \
-  --hf-subset "chart2text(cauldron)" \
-  --num-prompts 10
-```
-
-**`Aeala/ShareGPT_Vicuna_unfiltered`**
-
-```bash
-python3 vllm/benchmarks/benchmark_throughput.py \
-  --model Qwen/Qwen2-VL-7B-Instruct \
-  --backend vllm-chat \
-  --dataset-name hf \
-  --dataset-path Aeala/ShareGPT_Vicuna_unfiltered \
-  --hf-split train \
-  --num-prompts 10
-```
-
-**`AI-MO/aimo-validation-aime`**
-
-```bash
-python3 benchmarks/benchmark_throughput.py \
-  --model Qwen/QwQ-32B \
-  --backend vllm \
-  --dataset-name hf \
-  --dataset-path AI-MO/aimo-validation-aime \
-  --hf-split train \
-  --num-prompts 10
-```
-
-### Benchmark with LoRA Adapters
-
-``` bash
-# download dataset
-# wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
-python3 vllm/benchmarks/benchmark_throughput.py \
-  --model meta-llama/Llama-2-7b-hf \
-  --backend vllm \
-  --dataset_path <your data path>/ShareGPT_V3_unfiltered_cleaned_split.json \
-  --dataset_name sharegpt \
-  --num-prompts 10 \
-  --max-loras 2 \
-  --max-lora-rank 8 \
-  --enable-lora \
-  --lora-path yard1/llama-2-7b-sql-lora-test
-  ```

benchmarks/auto_tune.sh

@@ -1,212 +0,0 @@
-#!/bin/bash
-
-# This script aims to tune the best server parameter combinations to maximize throughput for given requirement. 
-# The current server parameter combination is  max_num_seqs and max_num_batched_tokens
-# It also supports additional requirement: e2e latency and prefix cache. 
-
-# Pre-requisite:
-# 1. Checkout to your branch, install/ update the correct running env. For TPU, activate conda env and install the corresponding torch, xla version. 
-# 2. If the model is customized, replace the MODEL's config with the customized config.
-# 3. Set variables (ALL REQUIRED)
-#   BASE: your directory for vllm repo
-#   MODEL: the model served by vllm
-#   DOWNLOAD_DIR: directory to download and load model weights.
-#   INPUT_LEN: request input len
-#   OUTPUT_LEN: request output len
-#   MIN_CACHE_HIT_PCT: prefix cache rate
-#   MAX_LATENCY_ALLOWED_MS: (e2e) latency requirement. If there's no latency requirement, set it to a large number like 1000000000
-# 4. Run the script, it might take a long time, you can use tmux to avoid the script stop if disconnection happens.
-# 5. The final result will be saved in RESULT file. 
-
-
-# Example use cases 
-# 1. Given input_len=1800, output_len=20, what's the best max_num_seqs and max_num_batched_tokens to get highest throughput?
-# Use INPUT_LEN=1800,  OUTPUT_LEN=20, MIN_CACHE_HIT_PCT=0, MAX_LATENCY_ALLOWED_MS=100000000000
-# 2. If we have latency requirement to be lower than 500ms, what's the best server parameter?
-# Use INPUT_LEN=1800,  OUTPUT_LEN=20, MIN_CACHE_HIT_PCT=0, MAX_LATENCY_ALLOWED_MS=500
-# 3. If we want to reach 60% prefix cache, what's the best server parameter? 
-# Use INPUT_LEN=1800,  OUTPUT_LEN=20, MIN_CACHE_HIT_PCT=60, MAX_LATENCY_ALLOWED_MS=500
-
-TAG=$(date +"%Y_%m_%d_%H_%M")
-BASE=""
-MODEL="meta-llama/Llama-3.1-8B-Instruct"
-DOWNLOAD_DIR=""
-INPUT_LEN=4000
-OUTPUT_LEN=16
-MIN_CACHE_HIT_PCT_PCT=0
-MAX_LATENCY_ALLOWED_MS=100000000000
-
-LOG_FOLDER="$BASE/auto-benchmark/$TAG"
-RESULT="$LOG_FOLDER/result.txt"
-
-echo "result file$ $RESULT"
-echo "model: $MODEL"
-echo
-
-rm -rf $LOG_FOLDER
-mkdir -p $LOG_FOLDER
-
-cd "$BASE/vllm"
-# create sonnet-4x.txt so that we can sample 2048 tokens for input
-echo "" > benchmarks/sonnet_4x.txt
-for _ in {1..4}
-do
-cat benchmarks/sonnet.txt >> benchmarks/sonnet_4x.txt
-done
-
-pip install datasets
-
-current_hash=$(git rev-parse HEAD)
-echo "hash:$current_hash" >> "$RESULT"
-echo "current_hash: $current_hash"
-
-best_throughput=0
-best_max_num_seqs=0
-best_num_batched_tokens=0
-best_goodput=0
-run_benchmark() {
-    local max_num_seqs=$1
-    local max_num_batched_tokens=$2
-    echo "max_num_seq: $max_num_seqs, max_num_batched_tokens: $max_num_batched_tokens"
-    local vllm_log="$LOG_FOLDER/vllm_log_${max_num_seqs}_${max_num_batched_tokens}.txt"
-    echo "vllm_log: $vllm_log"
-    echo
-    rm -f $vllm_log
-
-    # start the server
-    VLLM_USE_V1=1 VLLM_SERVER_DEV_MODE=1 vllm serve $MODEL \
-        --disable-log-requests \
-        --port 8004 \
-        --gpu-memory-utilization 0.98 \
-        --max-num-seqs $max_num_seqs \
-        --max-num-batched-tokens $max_num_batched_tokens \
-        --tensor-parallel-size 1 \
-        --enable-prefix-caching \
-        --load-format dummy \
-        --download-dir $DOWNLOAD_DIR \
-        --max-model-len $(( INPUT_LEN+OUTPUT_LEN )) > "$vllm_log" 2>&1 &
-    echo "wait for 10 minutes.."
-    echo
-    # wait for 10 minutes...
-    server_started=0
-    for i in {1..60}; do        
-        if grep -Fq "Application startup complete" "$vllm_log"; then
-            echo "Application started"
-            server_started=1
-            break
-        else
-            # echo "wait for 10 seconds..."
-            sleep 10
-        fi
-    done
- 
-    if (( ! server_started )); then
-        echo "server did not start within 10 minutes, terminate the benchmarking. Please check server log at $vllm_log"
-        echo "pkill -f vllm"
-        echo
-        pkill vllm
-        sleep 10
-        return 1
-    fi
-    
-    echo "run benchmark test..."
-    echo
-    meet_latency_requirement=0
-    # get a basic qps by using request-rate inf
-    bm_log="$LOG_FOLDER/bm_log_${max_num_seqs}_${max_num_batched_tokens}_requestrate_inf.txt"
-    prefix_len=$(( INPUT_LEN * MIN_CACHE_HIT_PCT / 100 ))
-    python benchmarks/benchmark_serving.py \
-        --backend vllm \
-        --model $MODEL  \
-        --dataset-name sonnet \
-        --dataset-path benchmarks/sonnet_4x.txt \
-        --sonnet-input-len $INPUT_LEN \
-        --sonnet-output-len $OUTPUT_LEN \
-        --ignore-eos \
-        --disable-tqdm \
-        --request-rate inf \
-        --percentile-metrics ttft,tpot,itl,e2el \
-        --goodput e2el:$MAX_LATENCY_ALLOWED_MS \
-        --num-prompts 100 \
-        --sonnet-prefix-len $prefix_len \
-        --port 8004 > "$bm_log"
-    through_put=$(grep "Request throughput (req/s):" "$bm_log" | sed 's/[^0-9.]//g')
-    e2el=$(grep "P99 E2EL (ms):" "$bm_log" | awk '{print $NF}')
-    goodput=$(grep "Request goodput (req/s):" "$bm_log" | sed 's/[^0-9.]//g')
-
-    if (( $(echo "$e2el <= $MAX_LATENCY_ALLOWED_MS" | bc -l) )); then
-        meet_latency_requirement=1
-    fi
-
-    if (( ! meet_latency_requirement )); then
-    # start from request-rate as int(through_put) + 1
-        request_rate=$((${through_put%.*} + 1))
-        while ((request_rate > 0)); do
-            # clear prefix cache
-            curl -X POST http://0.0.0.0:8004/reset_prefix_cache
-            sleep 5
-            bm_log="$LOG_FOLDER/bm_log_${max_num_seqs}_${max_num_batched_tokens}_requestrate_${request_rate}.txt"
-            python benchmarks/benchmark_serving.py \
-                --backend vllm \
-                --model $MODEL  \
-                --dataset-name sonnet \
-                --dataset-path benchmarks/sonnet_4x.txt \
-                --sonnet-input-len $INPUT_LEN \
-                --sonnet-output-len $OUTPUT_LEN \
-                --ignore_eos \
-                --disable-tqdm \
-                --request-rate $request_rate \
-                --percentile-metrics ttft,tpot,itl,e2el \
-                --goodput e2el:$MAX_LATENCY_ALLOWED_MS \
-                --num-prompts 100 \
-                --sonnet-prefix-len $prefix_len \
-                --port 8004 > "$bm_log"
-            through_put=$(grep "Request throughput (req/s):" "$bm_log" | sed 's/[^0-9.]//g')
-            e2el=$(grep "P99 E2EL (ms):" "$bm_log" | awk '{print $NF}')
-            goodput=$(grep "Request goodput (req/s):" "$bm_log" | sed 's/[^0-9.]//g')
-            if (( $(echo "$e2el <= $MAX_LATENCY_ALLOWED_MS" | bc -l) )); then
-                meet_latency_requirement=1
-                break
-            fi
-            request_rate=$((request_rate-1))
-        done
-    fi
-    # write the results and update the best result.
-    if ((meet_latency_requirement)); then
-        echo "max_num_seqs: $max_num_seqs, max_num_batched_tokens: $max_num_batched_tokens, request_rate: $request_rate, e2el: $e2el, through put: $through_put, goodput: $goodput"
-        echo "max_num_seqs: $max_num_seqs, max_num_batched_tokens: $max_num_batched_tokens, request_rate: $request_rate, e2el: $e2el, through put: $through_put, goodput: $goodput" >> "$RESULT"
-        if (( $(echo "$through_put > $best_throughput" | bc -l) )); then
-            best_throughput=$through_put
-            best_max_num_seqs=$max_num_seqs
-            best_num_batched_tokens=$max_num_batched_tokens
-            best_goodput=$goodput
-        fi
-    else
-        echo "max_num_seqs: $max_num_seqs, max_num_batched_tokens: $max_num_batched_tokens does not meet latency requirement ${MAX_LATENCY_ALLOWED_MS}"
-        echo "max_num_seqs: $max_num_seqs, max_num_batched_tokens: $max_num_batched_tokens does not meet latency requirement ${MAX_LATENCY_ALLOWED_MS}" >> "$RESULT"
-    fi
-
-    echo "best_max_num_seqs: $best_max_num_seqs, best_num_batched_tokens: $best_num_batched_tokens, best_throughput: $best_throughput"
-
-    echo "pkill -f vllm"
-    echo
-    pkill vllm
-    sleep 10
-    rm -f $vllm_log
-    printf '=%.0s' $(seq 1 20)
-    return 0
-}
-
-
-num_seqs_list="128 256"
-num_batched_tokens_list="512 1024 2048 4096"
-for num_seqs in $num_seqs_list; do
-    for num_batched_tokens in $num_batched_tokens_list; do
-        run_benchmark $num_seqs $num_batched_tokens
-        exit 0
-    done
-done
-echo "finish permutations"
-echo "best_max_num_seqs: $best_max_num_seqs, best_num_batched_tokens: $best_num_batched_tokens, best_throughput: $best_throughput"
-echo "best_max_num_seqs: $best_max_num_seqs, best_num_batched_tokens: $best_num_batched_tokens, best_throughput: $best_throughput" >> "$RESULT"
-

benchmarks/backend_request_func.py

@@ -1,13 +1,12 @@
 # SPDX-License-Identifier: Apache-2.0
 
-import io
 import json
 import os
 import sys
 import time
 import traceback
 from dataclasses import dataclass, field
-from typing import Optional, Union
+from typing import List, Optional, Union
 
 import aiohttp
 import huggingface_hub.constants
@@ -15,9 +14,6 @@ from tqdm.asyncio import tqdm
 from transformers import (AutoTokenizer, PreTrainedTokenizer,
                           PreTrainedTokenizerFast)
 
-# NOTE(simon): do not import vLLM here so the benchmark script
-# can run without vLLM installed.
-
 AIOHTTP_TIMEOUT = aiohttp.ClientTimeout(total=6 * 60 * 60)
 
 
@@ -29,11 +25,11 @@ class RequestFuncInput:
     output_len: int
     model: str
     model_name: Optional[str] = None
+    best_of: int = 1
     logprobs: Optional[int] = None
     extra_body: Optional[dict] = None
     multi_modal_content: Optional[dict] = None
     ignore_eos: bool = False
-    language: Optional[str] = None
 
 
 @dataclass
@@ -43,8 +39,8 @@ class RequestFuncOutput:
     latency: float = 0.0
     output_tokens: int = 0
     ttft: float = 0.0  # Time to first token
-    itl: list[float] = field(
-        default_factory=list)  # list of inter-token latencies
+    itl: List[float] = field(
+        default_factory=list)  # List of inter-token latencies
     tpot: float = 0.0  # avg next-token latencies
     prompt_len: int = 0
     error: str = ""
@@ -60,12 +56,13 @@ async def async_request_tgi(
     async with aiohttp.ClientSession(trust_env=True,
                                      timeout=AIOHTTP_TIMEOUT) as session:
         params = {
+            "best_of": request_func_input.best_of,
             "max_new_tokens": request_func_input.output_len,
             "do_sample": True,
             "temperature": 0.01,  # TGI does not accept 0.0 temperature.
             "top_p": 0.99,  # TGI does not accept 1.0 top_p.
             "truncate": request_func_input.prompt_len,
-            "ignore_eos_token": request_func_input.ignore_eos,
+            # TGI does not accept ignore_eos flag.
         }
         payload = {
             "inputs": request_func_input.prompt,
@@ -73,10 +70,6 @@ async def async_request_tgi(
         }
         output = RequestFuncOutput()
         output.prompt_len = request_func_input.prompt_len
-        if request_func_input.ignore_eos:
-            output.output_tokens = request_func_input.output_len
-        else:
-            output.output_tokens = None
 
         ttft = 0.0
         st = time.perf_counter()
@@ -135,6 +128,7 @@ async def async_request_trt_llm(
 
     async with aiohttp.ClientSession(trust_env=True,
                                      timeout=AIOHTTP_TIMEOUT) as session:
+        assert request_func_input.best_of == 1
         payload = {
             "accumulate_tokens": True,
             "text_input": request_func_input.prompt,
@@ -199,9 +193,9 @@ async def async_request_deepspeed_mii(
 ) -> RequestFuncOutput:
     async with aiohttp.ClientSession(trust_env=True,
                                      timeout=AIOHTTP_TIMEOUT) as session:
+        assert request_func_input.best_of == 1
 
         payload = {
-            "model": request_func_input.model,
             "prompt": request_func_input.prompt,
             "max_tokens": request_func_input.output_len,
             "temperature": 0.01,  # deepspeed-mii does not accept 0.0 temp.
@@ -222,15 +216,7 @@ async def async_request_deepspeed_mii(
                 if response.status == 200:
                     parsed_resp = await response.json()
                     output.latency = time.perf_counter() - st
-                    if "choices" in parsed_resp:
-                        output.generated_text = parsed_resp["choices"][0][
-                            "text"]
-                    elif "text" in parsed_resp:
-                        output.generated_text = parsed_resp["text"][0]
-                    else:
-                        output.error = ("Unexpected response format: "
-                                        "neither 'choices' nor 'text' found")
-                        output.success = False
+                    output.generated_text = parsed_resp["text"][0]
                     output.success = True
                 else:
                     output.error = response.reason or ""
@@ -261,7 +247,7 @@ async def async_request_openai_completions(
                 if request_func_input.model_name else request_func_input.model,
             "prompt": request_func_input.prompt,
             "temperature": 0.0,
-            "repetition_penalty": 1.0,
+            "best_of": request_func_input.best_of,
             "max_tokens": request_func_input.output_len,
             "logprobs": request_func_input.logprobs,
             "stream": True,
@@ -350,7 +336,7 @@ async def async_request_openai_chat_completions(
 ) -> RequestFuncOutput:
     api_url = request_func_input.api_url
     assert api_url.endswith(
-        ("chat/completions", "profile")
+        "chat/completions"
     ), "OpenAI Chat Completions API URL must end with 'chat/completions'."
 
     async with aiohttp.ClientSession(trust_env=True,
@@ -440,125 +426,16 @@ async def async_request_openai_chat_completions(
     return output
 
 
-async def async_request_openai_audio(
-    request_func_input: RequestFuncInput,
-    pbar: Optional[tqdm] = None,
-) -> RequestFuncOutput:
-    # Lazy import without PlaceholderModule to avoid vllm dep.
-    import soundfile
-    api_url = request_func_input.api_url
-    assert api_url.endswith(
-        ("transcriptions", "translations"
-         )), "OpenAI Chat Completions API URL must end with 'transcriptions' "
-    "or `translations`."
-
-    async with aiohttp.ClientSession(trust_env=True,
-                                     timeout=AIOHTTP_TIMEOUT) as session:
-        content = [{"type": "text", "text": request_func_input.prompt}]
-        payload = {
-            "model": request_func_input.model_name \
-                if request_func_input.model_name else request_func_input.model,
-            "temperature": 0.0,
-            "max_completion_tokens": request_func_input.output_len,
-            "stream": True,
-            "language": "en",
-            # Flattened due to multipart/form-data
-            "stream_include_usage": True,
-            "stream_continuous_usage_stats": True
-        }
-        if request_func_input.extra_body:
-            payload.update(request_func_input.extra_body)
-        headers = {
-            "Authorization": f"Bearer {os.environ.get('OPENAI_API_KEY')}",
-        }
-
-        # Send audio file
-        def to_bytes(y, sr):
-            buffer = io.BytesIO()
-            soundfile.write(buffer, y, sr, format="WAV")
-            buffer.seek(0)
-            return buffer
-
-        with to_bytes(*request_func_input.multi_modal_content['audio']) as f:
-            form = aiohttp.FormData()
-            form.add_field('file', f, content_type='audio/wav')
-            for key, value in payload.items():
-                form.add_field(key, str(value))
-
-            output = RequestFuncOutput()
-            output.prompt_len = request_func_input.prompt_len
-
-            generated_text = ""
-            ttft = 0.0
-            st = time.perf_counter()
-            most_recent_timestamp = st
-            try:
-                async with session.post(url=api_url,
-                                        data=form,
-                                        headers=headers) as response:
-                    if response.status == 200:
-                        async for chunk_bytes in response.content:
-                            chunk_bytes = chunk_bytes.strip()
-                            if not chunk_bytes:
-                                continue
-
-                            chunk = chunk_bytes.decode("utf-8").removeprefix(
-                                "data: ")
-                            if chunk != "[DONE]":
-                                timestamp = time.perf_counter()
-                                data = json.loads(chunk)
-
-                                if choices := data.get("choices"):
-                                    content = choices[0]["delta"].get(
-                                        "content")
-                                    # First token
-                                    if ttft == 0.0:
-                                        ttft = timestamp - st
-                                        output.ttft = ttft
-
-                                    # Decoding phase
-                                    else:
-                                        output.itl.append(
-                                            timestamp - most_recent_timestamp)
-
-                                    generated_text += content or ""
-                                elif usage := data.get("usage"):
-                                    output.output_tokens = usage.get(
-                                        "completion_tokens")
-
-                                most_recent_timestamp = timestamp
-
-                        output.generated_text = generated_text
-                        output.success = True
-                        output.latency = most_recent_timestamp - st
-                    else:
-                        output.error = response.reason or ""
-                        output.success = False
-            except Exception:
-                output.success = False
-                exc_info = sys.exc_info()
-                output.error = "".join(traceback.format_exception(*exc_info))
-
-        if pbar:
-            pbar.update(1)
-        return output
-
-
 def get_model(pretrained_model_name_or_path: str) -> str:
     if os.getenv('VLLM_USE_MODELSCOPE', 'False').lower() == 'true':
         from modelscope import snapshot_download
 
-        from vllm.model_executor.model_loader.weight_utils import get_lock
+        model_path = snapshot_download(
+            model_id=pretrained_model_name_or_path,
+            local_files_only=huggingface_hub.constants.HF_HUB_OFFLINE,
+            ignore_file_pattern=[".*.pt", ".*.safetensors", ".*.bin"])
 
-        # Use file lock to prevent multiple processes from
-        # downloading the same model weights at the same time.
-        with get_lock(pretrained_model_name_or_path):
-            model_path = snapshot_download(
-                model_id=pretrained_model_name_or_path,
-                local_files_only=huggingface_hub.constants.HF_HUB_OFFLINE,
-                ignore_file_pattern=[".*.pt", ".*.safetensors", ".*.bin"])
-
-            return model_path
+        return model_path
     return pretrained_model_name_or_path
 
 
@@ -601,14 +478,7 @@ ASYNC_REQUEST_FUNCS = {
     "deepspeed-mii": async_request_deepspeed_mii,
     "openai": async_request_openai_completions,
     "openai-chat": async_request_openai_chat_completions,
-    "openai-audio": async_request_openai_audio,
     "tensorrt-llm": async_request_trt_llm,
     "scalellm": async_request_openai_completions,
     "sglang": async_request_openai_completions,
 }
-
-OPENAI_COMPATIBLE_BACKENDS = [
-    k for k, v in ASYNC_REQUEST_FUNCS.items()
-    if v in (async_request_openai_completions,
-             async_request_openai_chat_completions)
-]

benchmarks/benchmark_guided.py

@@ -0,0 +1,495 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Benchmark guided decoding throughput."""
+import argparse
+import dataclasses
+import json
+import os
+import random
+import time
+from typing import List
+
+import datasets
+import pandas as pd
+import uvloop
+from transformers import AutoTokenizer, PreTrainedTokenizerBase
+
+from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs
+from vllm.entrypoints.openai.api_server import (
+    build_async_engine_client_from_engine_args)
+from vllm.sampling_params import GuidedDecodingParams
+from vllm.utils import FlexibleArgumentParser, merge_async_iterators
+
+
+@dataclasses.dataclass
+class SampleRequest:
+    """A class representing a single inference request for benchmarking.
+
+    Attributes:
+        prompt: The input text prompt for the model.
+        multi_modal_data: Optional dictionary containing multi-modal data (e.g.
+            images).
+        prompt_len: The length of the prompt in tokens.
+        expected_output_len: The expected length of the output in tokens.
+    """
+    prompt: str
+    prompt_len: int
+    expected_output_len: int
+    schema: dict
+    structure_type: str = 'json'
+    completion: str = None
+
+
+def run_vllm(requests: List[SampleRequest],
+             engine_args: EngineArgs,
+             n: int,
+             guided_decoding_rate: float = 1.0,
+             warmup: bool = False) -> float:
+    from vllm import LLM, SamplingParams
+    llm = LLM(**vars(engine_args))
+
+    # Add the requests to the engine.
+    prompts: List[str] = []
+    sampling_params: List[SamplingParams] = []
+    # create a list containing random selected true or false
+    guided_decoding_req_idx = random.sample(
+        range(len(requests)), int(len(requests) * guided_decoding_rate))
+
+    if warmup:
+        print(">>>>> Running warmup prompt, for the first 5")
+        # We setup the first 5 requests to warmup FSM
+        # if using xgrammar dataset, we will skip warmup
+        warmup_requests = requests[:5]
+        for i, request in enumerate(warmup_requests):
+            prompts.append(request.prompt)
+            sampling_params.append(
+                SamplingParams(
+                    n=n,
+                    temperature=1.0,
+                    top_p=1.0,
+                    ignore_eos=True,
+                    max_tokens=request.expected_output_len,
+                    guided_decoding=GuidedDecodingParams(json=request.schema)
+                    if guided_decoding_rate > 0 else None,
+                ))
+        llm.generate(prompts, sampling_params, use_tqdm=False)
+
+    print(">>>>> Benchmark started...")
+    prompts = []
+    sampling_params = []
+    for i, request in enumerate(requests):
+        prompts.append(request.prompt)
+        sampling_params.append(
+            SamplingParams(
+                n=n,
+                temperature=1.0,
+                top_p=1.0,
+                ignore_eos=True,
+                max_tokens=request.expected_output_len,
+                guided_decoding=GuidedDecodingParams(
+                    **{request.structure_type: request.schema})
+                if i in guided_decoding_req_idx else None,
+            ))
+
+    start = time.perf_counter()
+    outputs = llm.generate(prompts, sampling_params, use_tqdm=False)
+    ret = []
+    for output, request in zip(outputs, requests):
+        generated_text = output.outputs[0].text
+        ret.append({
+            "generated": generated_text,
+            "expected": request.completion
+        })
+    end = time.perf_counter()
+    return end - start, ret
+
+
+async def run_vllm_async(
+        requests: List[SampleRequest],
+        engine_args: AsyncEngineArgs,
+        n: int,
+        guided_decoding_rate: float = 1.0,
+        warmup: bool = False,
+        disable_frontend_multiprocessing: bool = False) -> float:
+    from vllm import SamplingParams
+
+    async with build_async_engine_client_from_engine_args(
+            engine_args, disable_frontend_multiprocessing) as llm:
+
+        # Add the requests to the engine.
+        prompts: List[str] = []
+        sampling_params: List[SamplingParams] = []
+        guided_decoding_req_idx = random.sample(
+            range(len(requests)), int(len(requests) * guided_decoding_rate))
+
+        if warmup:
+            print(">>>>>> Running warmup prompt, for the first 5")
+            # We setup the first 5 requests to warmup FSM
+            # if using xgrammar dataset, we will skip warmup
+            warmup_requests = requests[:5]
+            for i, request in enumerate(warmup_requests):
+                prompts.append(request.prompt)
+                sampling_params.append(
+                    SamplingParams(
+                        n=n,
+                        temperature=1.0,
+                        top_p=1.0,
+                        ignore_eos=True,
+                        max_tokens=request.expected_output_len,
+                        guided_decoding=GuidedDecodingParams(
+                            json=request.schema)
+                        if guided_decoding_rate > 0 else None,
+                    ))
+            generators = []
+            for i, (prompt, sp) in enumerate(zip(prompts, sampling_params)):
+                generator = llm.generate(prompt, sp, request_id=f"test{i}")
+                generators.append(generator)
+            all_gens = merge_async_iterators(*generators)
+            async for i, res in all_gens:
+                pass
+
+        print(">>>>> Benchmark started...")
+        prompts = []
+        sampling_params = []
+        for i, request in enumerate(requests):
+            prompts.append(request.prompt)
+            sampling_params.append(
+                SamplingParams(
+                    n=n,
+                    temperature=1.0,
+                    top_p=1.0,
+                    ignore_eos=True,
+                    max_tokens=request.expected_output_len,
+                    guided_decoding=GuidedDecodingParams(json=request.schema)
+                    if i in guided_decoding_req_idx else None,
+                ))
+
+        generators = []
+        start_time = []
+        latencies = []
+        start = time.perf_counter()
+        for i, (prompt, sp) in enumerate(zip(prompts, sampling_params)):
+            generator = llm.generate(prompt, sp, request_id=f"test{i}")
+            generators.append(generator)
+            start_time.append(time.perf_counter())
+            latencies.append([])
+        all_gens = merge_async_iterators(*generators)
+        generated_texts = [''] * len(requests)
+        async for i, res in all_gens:
+            generated_texts[i] = res.outputs[0].text
+            lat = time.perf_counter() - start_time[i]
+            latencies[i].append(lat)
+        ret = [{
+            'generated': gt,
+            'expected': req.completion
+        } for gt, req in zip(generated_texts, requests)]
+        end = time.perf_counter()
+        first_latency = pd.Series([lat[0] * 1000 for lat in latencies])
+        next_latency = pd.Series([(lat[-1] - lat[0]) / len(lat[1:]) * 1000
+                                  for lat in latencies])
+        return end - start, ret, (first_latency, next_latency)
+
+
+def sample_requests(tokenizer: PreTrainedTokenizerBase,
+                    args: argparse.Namespace) -> List[SampleRequest]:
+    if args.dataset == 'json':
+        if args.json_schema_path is None:
+            dir_path = os.path.dirname(os.path.realpath(__file__))
+            args.json_schema_path = os.path.join(dir_path,
+                                                 "structured_schemas",
+                                                 "structured_schema_1.json")
+        with open(args.json_schema_path) as f:
+            schema = json.load(f)
+        prompt = f"Generate an example of a user profile given the following schema: {json.dumps(schema)}"  # noqa: E501
+        input_len = len(tokenizer(prompt).input_ids)
+        print(f"Input length of the prompt: {input_len} tokens")
+        requests = [
+            SampleRequest(prompt=prompt,
+                          prompt_len=input_len,
+                          expected_output_len=args.output_len,
+                          schema=schema,
+                          structure_type=args.structure_type)
+            for _ in range(args.num_prompts)
+        ]
+
+    elif args.dataset == "grammar":
+        schema = """
+            ?start: select_statement
+
+            ?select_statement: "SELECT " column_list " FROM " table_name
+
+            ?column_list: column_name ("," column_name)*
+
+            ?table_name: identifier
+
+            ?column_name: identifier
+
+            ?identifier: /[a-zA-Z_][a-zA-Z0-9_]*/
+        """
+        prompt = "Generate an SQL query to show the 'username' \
+            and 'email' from the 'users' table."
+
+        input_len = len(tokenizer(prompt).input_ids)
+        print(f"Input length of the prompt: {input_len} tokens")
+        requests = [
+            SampleRequest(prompt=prompt,
+                          prompt_len=input_len,
+                          expected_output_len=args.output_len,
+                          schema=schema,
+                          structure_type=args.structure_type)
+            for _ in range(args.num_prompts)
+        ]
+
+    elif args.dataset == "regex":
+        regex = r"\w+@\w+\.com\n"
+        args.regex = regex
+        prompt = "Generate an email address for Alan Turing, \
+            who works in Enigma. End in .com and new line. \
+                Example result: alan.turing@enigma.com\n"
+
+        input_len = len(tokenizer(prompt).input_ids)
+        print(f"Input length of the prompt: {input_len} tokens")
+        requests = [
+            SampleRequest(prompt=prompt,
+                          prompt_len=input_len,
+                          expected_output_len=args.output_len,
+                          schema=regex,
+                          structure_type=args.structure_type)
+            for _ in range(args.num_prompts)
+        ]
+
+    elif args.dataset == "choice":
+        choice = ["Positive", "Negative"]
+        args.choice = choice
+        prompt = "Classify this sentiment: vLLM is wonderful!"
+        input_len = len(tokenizer(prompt).input_ids)
+        print(f"Input length of the prompt: {input_len} tokens")
+        requests = [
+            SampleRequest(prompt=prompt,
+                          prompt_len=input_len,
+                          expected_output_len=args.output_len,
+                          schema=choice,
+                          structure_type=args.structure_type)
+            for _ in range(args.num_prompts)
+        ]
+
+    elif args.dataset == "xgrammar_bench":
+        args.warmup = False
+        requests: List[SampleRequest] = []
+        dataset = datasets.load_dataset("NousResearch/json-mode-eval",
+                                        split="train")
+        print(f"dataset has {len(dataset)} entries")
+        len_dataset = len(dataset)
+        for data_point_idx in range(args.num_prompts):
+            idx = data_point_idx
+            while idx >= len_dataset:
+                idx -= len_dataset
+            schema = dataset["schema"][idx]
+            prompt = tokenizer.apply_chat_template(dataset["prompt"][idx],
+                                                   tokenize=False)
+            input_len = len(tokenizer(prompt).input_ids)
+            completion = dataset["completion"][idx]
+
+            requests.append(
+                SampleRequest(prompt=prompt,
+                              prompt_len=input_len,
+                              expected_output_len=args.output_len,
+                              schema=schema,
+                              completion=completion))
+
+    return requests
+
+
+def evaluate(ret, args):
+
+    def _eval_correctness_json(expected, actual):
+        # extract json string from string using regex
+        import re
+        actual = actual.replace('\n', '').replace(' ', '').strip()
+        try:
+            actual = re.search(r'\{.*\}', actual).group()
+            actual = json.loads(actual)
+        except Exception:
+            return False
+
+        return True
+
+    def _eval_correctness_choice(expected, actual):
+        return actual in args.choice
+
+    def _eval_correctness_regex(expected, actual):
+        import re
+        return re.match(args.regex, actual) is not None
+
+    def _eval_correctness(expected, actual):
+        if args.structure_type == 'json':
+            return _eval_correctness_json(expected, actual)
+        elif args.structure_type == 'regex':
+            return _eval_correctness_regex(expected, actual)
+        elif args.structure_type == 'choice':
+            return _eval_correctness_choice(expected, actual)
+        else:
+            return None
+
+    scores = []
+    for res in ret:
+        score = _eval_correctness(res['expected'], res['generated'])
+        res['correctness'] = score
+        scores.append(score)
+
+    not_none_scores = [score for score in scores if score is not None]
+
+    return (sum(not_none_scores) / len(not_none_scores) *
+            100) if len(not_none_scores) > 0 else None
+
+
+def main(args: argparse.Namespace):
+    print(args)
+    random.seed(args.seed)
+
+    # async engine is working for 'regex', 'choice' and 'grammar'
+    if args.dataset == 'grammar':
+        args.structure_type = 'grammar'
+        args.async_engine = False
+    elif args.dataset == 'regex':
+        args.structure_type = 'regex'
+        args.async_engine = False
+    elif args.dataset == 'choice':
+        args.structure_type = 'choice'
+        args.async_engine = False
+    else:
+        args.structure_type = 'json'
+
+    if args.no_guided_decoding:
+        args.guided_decoding_ratio = 0
+    if args.save_results:
+        result_file_name = f'{args.guided_decoding_ratio}guided'
+        result_file_name += f"_{args.model.split('/')[-1]}"
+        result_file_name += f"_{args.dataset}"
+        result_file_name += f"_{args.num_prompts}"
+        result_file_name += f"_out{args.output_len}"
+        result_file_name += f"_async{args.async_engine}"
+        result_file_name += f"_warmup{args.warmup}"
+        result_file_name += f"_chunkedprefill{args.enable_chunked_prefill}"
+        result_file_name += ".txt"
+    else:
+        result_file_name = None
+
+    # Synthesize a prompt with the given input length.
+    tokenizer = AutoTokenizer.from_pretrained(
+        args.tokenizer, trust_remote_code=args.trust_remote_code)
+    requests = sample_requests(tokenizer, args)
+
+    if args.async_engine:
+        engine_args = AsyncEngineArgs.from_cli_args(args)
+        elapsed_time, ret, (first_latency, next_latency) = uvloop.run(
+            run_vllm_async(requests, engine_args, args.n,
+                           args.guided_decoding_ratio, args.warmup,
+                           args.disable_frontend_multiprocessing))
+    else:
+        engine_args = EngineArgs.from_cli_args(args)
+        elapsed_time, ret = run_vllm(requests, engine_args, args.n,
+                                     args.guided_decoding_ratio, args.warmup)
+        first_latency, next_latency = None, None
+
+    score = evaluate(ret, args)
+    total_num_tokens = sum(request.prompt_len + request.expected_output_len
+                           for request in requests)
+    total_output_tokens = sum(request.expected_output_len
+                              for request in requests)
+    if first_latency is not None:
+        latency_breakdown = "\nFirst token latency(msecs):\n"
+        latency_breakdown += f"{first_latency.describe()}"
+        latency_breakdown += "\nNext token latency(msecs):\n"
+        latency_breakdown += f"{next_latency.describe()}"
+    print(
+        f"Throughput: {len(requests) / elapsed_time:.2f} requests/s, "
+        f"{total_num_tokens / elapsed_time:.2f} total tokens/s, "
+        f"{total_output_tokens / elapsed_time:.2f} output tokens/s",
+        f"Correct rate is {score} %",
+        f"{latency_breakdown if first_latency is not None else ''}")
+
+    # Output JSON results if specified
+    if args.output_json or result_file_name:
+        results = {
+            "elapsed_time": elapsed_time,
+            "num_requests": len(requests),
+            "total_num_tokens": total_num_tokens,
+            "total_output_tokens": total_output_tokens,
+            "requests_per_second": len(requests) / elapsed_time,
+            "tokens_per_second": f"{total_num_tokens / elapsed_time:.2f}",
+            "output_tokens_per_second":
+            f"{total_output_tokens / elapsed_time:.2f}",
+            "correct_rate(%)": score
+        }
+        results = {"outputs": ret, **results}
+        if first_latency is not None:
+            results["first_token_latency(msecs)"] = first_latency.describe(
+            ).to_dict()
+            results["next_token_latency(msecs)"] = next_latency.describe(
+            ).to_dict()
+        if args.output_json:
+            with open(args.output_json, "w") as f:
+                json.dump(results, f, indent=4)
+        elif result_file_name:
+            with open(result_file_name, "w") as f:
+                json.dump(results, f, indent=4)
+
+
+if __name__ == "__main__":
+    parser = FlexibleArgumentParser(description="Benchmark guided decoding.")
+    parser = AsyncEngineArgs.add_cli_args(parser)
+
+    parser.add_argument("--output-len",
+                        type=int,
+                        default=512,
+                        help="Output length for each request. Overrides the "
+                        "output length from the dataset.")
+    parser.add_argument(
+        "--dataset",
+        default='json',
+        choices=['json', 'grammar', 'regex', 'choice', 'xgrammar_bench'])
+    parser.add_argument("--json_schema_path",
+                        type=str,
+                        default=None,
+                        help="Path to json schema.")
+    parser.add_argument("--n",
+                        type=int,
+                        default=1,
+                        help="Number of generated sequences per prompt.")
+    parser.add_argument("--num-prompts",
+                        type=int,
+                        default=10,
+                        help="Number of prompts to process.")
+    parser.add_argument(
+        '--output-json',
+        type=str,
+        default=None,
+        help='Path to save the throughput results in JSON format.')
+    parser.add_argument("--async-engine",
+                        action='store_true',
+                        default=False,
+                        help="Use vLLM async engine rather than LLM class.")
+    parser.add_argument("--no-guided-decoding",
+                        action='store_true',
+                        default=False,
+                        help="Whether to disable JSON decoding or not.")
+    parser.add_argument("--guided-decoding-ratio",
+                        type=float,
+                        default=1.0,
+                        help="Ratio of Guided Decoding requests")
+    parser.add_argument("--disable-frontend-multiprocessing",
+                        action='store_true',
+                        default=False,
+                        help="Disable decoupled async engine frontend.")
+    parser.add_argument("--warmup",
+                        action="store_true",
+                        default=False,
+                        help="Run warmup prompts before benchmark.")
+    parser.add_argument("--save-results",
+                        action="store_true",
+                        default=False,
+                        help="save output results.")
+    args = parser.parse_args()
+    if args.tokenizer is None:
+        args.tokenizer = args.model
+    main(args)

benchmarks/benchmark_latency.py

@@ -1,17 +1,14 @@
 # SPDX-License-Identifier: Apache-2.0
 """Benchmark the latency of processing a single batch of requests."""
-
 import argparse
 import dataclasses
 import json
-import os
 import time
 from pathlib import Path
-from typing import Any, Optional
+from typing import List, Optional
 
 import numpy as np
 import torch
-from benchmark_utils import convert_to_pytorch_benchmark_format, write_to_json
 from tqdm import tqdm
 
 from vllm import LLM, SamplingParams
@@ -21,18 +18,6 @@ from vllm.sampling_params import BeamSearchParams
 from vllm.utils import FlexibleArgumentParser
 
 
-def save_to_pytorch_benchmark_format(args: argparse.Namespace,
-                                     results: dict[str, Any]) -> None:
-    pt_records = convert_to_pytorch_benchmark_format(
-        args=args,
-        metrics={"latency": results["latencies"]},
-        extra_info={k: results[k]
-                    for k in ["avg_latency", "percentiles"]})
-    if pt_records:
-        pt_file = f"{os.path.splitext(args.output_json)[0]}.pytorch.json"
-        write_to_json(pt_file, pt_records)
-
-
 def main(args: argparse.Namespace):
     print(args)
 
@@ -41,10 +26,6 @@ def main(args: argparse.Namespace):
     # NOTE(woosuk): If the request cannot be processed in a single batch,
     # the engine will automatically process the request in multiple batches.
     llm = LLM(**dataclasses.asdict(engine_args))
-    assert llm.llm_engine.model_config.max_model_len >= (
-        args.input_len +
-        args.output_len), ("Please ensure that max_model_len is greater than"
-                           " the sum of input_len and output_len.")
 
     sampling_params = SamplingParams(
         n=args.n,
@@ -52,13 +33,12 @@ def main(args: argparse.Namespace):
         top_p=1.0,
         ignore_eos=True,
         max_tokens=args.output_len,
-        detokenize=not args.disable_detokenize,
     )
     print(sampling_params)
     dummy_prompt_token_ids = np.random.randint(10000,
                                                size=(args.batch_size,
                                                      args.input_len))
-    dummy_prompts: list[PromptType] = [{
+    dummy_prompts: List[PromptType] = [{
         "prompt_token_ids": batch
     } for batch in dummy_prompt_token_ids.tolist()]
 
@@ -74,8 +54,7 @@ def main(args: argparse.Namespace):
                     beam_width=args.n,
                     max_tokens=args.output_len,
                     ignore_eos=True,
-                ),
-            )
+                ))
 
     def run_to_completion(profile_dir: Optional[str] = None):
         if profile_dir:
@@ -85,8 +64,7 @@ def main(args: argparse.Namespace):
                         torch.profiler.ProfilerActivity.CUDA,
                     ],
                     on_trace_ready=torch.profiler.tensorboard_trace_handler(
-                        str(profile_dir)),
-            ) as p:
+                        str(profile_dir))) as p:
                 llm_generate()
             print(p.key_averages().table(sort_by="self_cuda_time_total"))
         else:
@@ -103,8 +81,9 @@ def main(args: argparse.Namespace):
     if args.profile:
         profile_dir = args.profile_result_dir
         if not profile_dir:
-            profile_dir = (Path(".") / "vllm_benchmark_result" /
-                           f"latency_result_{time.time()}")
+            profile_dir = Path(
+                "."
+            ) / "vllm_benchmark_result" / f"latency_result_{time.time()}"
         print(f"Profiling (results will be saved to '{profile_dir}')...")
         run_to_completion(profile_dir=profile_dir)
         return
@@ -116,9 +95,9 @@ def main(args: argparse.Namespace):
     latencies = np.array(latencies)
     percentages = [10, 25, 50, 75, 90, 99]
     percentiles = np.percentile(latencies, percentages)
-    print(f"Avg latency: {np.mean(latencies)} seconds")
+    print(f'Avg latency: {np.mean(latencies)} seconds')
     for percentage, percentile in zip(percentages, percentiles):
-        print(f"{percentage}% percentile latency: {percentile} seconds")
+        print(f'{percentage}% percentile latency: {percentile} seconds')
 
     # Output JSON results if specified
     if args.output_json:
@@ -129,57 +108,43 @@ def main(args: argparse.Namespace):
         }
         with open(args.output_json, "w") as f:
             json.dump(results, f, indent=4)
-        save_to_pytorch_benchmark_format(args, results)
 
 
-if __name__ == "__main__":
+if __name__ == '__main__':
     parser = FlexibleArgumentParser(
-        description="Benchmark the latency of processing a single batch of "
-        "requests till completion.")
-    parser.add_argument("--input-len", type=int, default=32)
-    parser.add_argument("--output-len", type=int, default=128)
-    parser.add_argument("--batch-size", type=int, default=8)
-    parser.add_argument(
-        "--n",
-        type=int,
-        default=1,
-        help="Number of generated sequences per prompt.",
-    )
-    parser.add_argument("--use-beam-search", action="store_true")
-    parser.add_argument(
-        "--num-iters-warmup",
-        type=int,
-        default=10,
-        help="Number of iterations to run for warmup.",
-    )
-    parser.add_argument("--num-iters",
+        description='Benchmark the latency of processing a single batch of '
+        'requests till completion.')
+    parser.add_argument('--input-len', type=int, default=32)
+    parser.add_argument('--output-len', type=int, default=128)
+    parser.add_argument('--batch-size', type=int, default=8)
+    parser.add_argument('--n',
+                        type=int,
+                        default=1,
+                        help='Number of generated sequences per prompt.')
+    parser.add_argument('--use-beam-search', action='store_true')
+    parser.add_argument('--num-iters-warmup',
+                        type=int,
+                        default=10,
+                        help='Number of iterations to run for warmup.')
+    parser.add_argument('--num-iters',
                         type=int,
                         default=30,
-                        help="Number of iterations to run.")
+                        help='Number of iterations to run.')
     parser.add_argument(
-        "--profile",
-        action="store_true",
-        help="profile the generation process of a single batch",
-    )
+        '--profile',
+        action='store_true',
+        help='profile the generation process of a single batch')
     parser.add_argument(
-        "--profile-result-dir",
+        '--profile-result-dir',
         type=str,
         default=None,
-        help=("path to save the pytorch profiler output. Can be visualized "
-              "with ui.perfetto.dev or Tensorboard."),
-    )
+        help=('path to save the pytorch profiler output. Can be visualized '
+              'with ui.perfetto.dev or Tensorboard.'))
     parser.add_argument(
-        "--output-json",
+        '--output-json',
         type=str,
         default=None,
-        help="Path to save the latency results in JSON format.",
-    )
-    parser.add_argument(
-        "--disable-detokenize",
-        action="store_true",
-        help=("Do not detokenize responses (i.e. do not include "
-              "detokenization time in the latency measurement)"),
-    )
+        help='Path to save the latency results in JSON format.')
 
     parser = EngineArgs.add_cli_args(parser)
     args = parser.parse_args()

benchmarks/benchmark_prefix_caching.py

@@ -31,7 +31,7 @@ import dataclasses
 import json
 import random
 import time
-from typing import Optional
+from typing import List, Optional, Tuple
 
 from transformers import PreTrainedTokenizerBase
 
@@ -63,25 +63,23 @@ class Request:
     output_len: int
 
 
-def sample_tokens(tokenizer: PreTrainedTokenizerBase,
-                  length: int) -> list[int]:
+def sample_tokens(tokenizer: PreTrainedTokenizerBase, length: int) -> str:
     vocab = tokenizer.get_vocab()
-    all_special_ids = set(tokenizer.all_special_ids)
-
     # Remove the special tokens.
-    return random.choices(
-        [v for k, v in vocab.items() if k not in all_special_ids],
-        k=length,
-    )
+    vocab = {
+        k: v
+        for k, v in vocab.items() if k not in tokenizer.all_special_ids
+    }
+    return random.choices(list(vocab.values()), k=length)
 
 
 def sample_requests_from_dataset(
     dataset_path: str,
     num_requests: int,
     tokenizer: PreTrainedTokenizerBase,
-    input_length_range: tuple[int, int],
+    input_length_range: Tuple[int, int],
     fixed_output_len: Optional[int],
-) -> list[Request]:
+) -> List[Request]:
     if fixed_output_len is not None and fixed_output_len < 4:
         raise ValueError("output_len too small")
 
@@ -101,7 +99,7 @@ def sample_requests_from_dataset(
     assert min_len >= 0 and max_len >= min_len, "input_length_range too small"
 
     # Filter out sequences that are too long or too short
-    filtered_requests: list[Request] = []
+    filtered_requests: List[Request] = []
 
     for i in range(len(dataset)):
         if len(filtered_requests) == num_requests:
@@ -124,10 +122,10 @@ def sample_requests_from_dataset(
 def sample_requests_from_random(
     num_requests: int,
     tokenizer: PreTrainedTokenizerBase,
-    input_length_range: tuple[int, int],
+    input_length_range: Tuple[int, int],
     fixed_output_len: Optional[int],
     prefix_len: int,
-) -> list[Request]:
+) -> List[Request]:
 
     requests = []
     prefix_token_ids = sample_tokens(tokenizer, prefix_len)
@@ -146,9 +144,9 @@ def sample_requests_from_random(
     return requests
 
 
-def repeat_and_sort_requests(requests: list[Request],
+def repeat_and_sort_requests(requests: List[Request],
                              repeat_count: int,
-                             sort: bool = False) -> list[str]:
+                             sort: bool = False) -> List[str]:
     repeated_requests = requests * repeat_count
     if sort:
         repeated_requests.sort(key=lambda x: x[1])
@@ -196,9 +194,7 @@ def main(args):
 
     llm = LLM(**dataclasses.asdict(engine_args))
 
-    sampling_params = SamplingParams(temperature=0,
-                                     max_tokens=args.output_len,
-                                     detokenize=not args.disable_detokenize)
+    sampling_params = SamplingParams(temperature=0, max_tokens=args.output_len)
 
     print("Testing filtered requests")
     prompts = repeat_and_sort_requests(filtered_requests,
@@ -247,12 +243,6 @@ if __name__ == "__main__":
         "subtract this length when filtering prompts. Only used "
         "when dataset-path is not provided.",
     )
-    parser.add_argument(
-        '--disable-detokenize',
-        action='store_true',
-        help=("Do not detokenize responses (i.e. do not include "
-              "detokenization time in the latency measurement)"),
-    )
 
     parser = EngineArgs.add_cli_args(parser)
     args = parser.parse_args()

benchmarks/benchmark_prioritization.py

@@ -5,7 +5,7 @@ import dataclasses
 import json
 import random
 import time
-from typing import Optional
+from typing import List, Optional, Tuple
 
 from transformers import AutoTokenizer, PreTrainedTokenizerBase
 
@@ -13,17 +13,12 @@ from vllm.engine.arg_utils import EngineArgs
 from vllm.utils import FlexibleArgumentParser
 
 
-#Select a equi-probable random priority
-def get_random_flag():
-    return 0 if random.random() < 0.5 else 1
-
-
 def sample_requests(
     dataset_path: str,
     num_requests: int,
     tokenizer: PreTrainedTokenizerBase,
     fixed_output_len: Optional[int],
-) -> list[tuple[str, int, int, int]]:
+) -> List[Tuple[str, int, int]]:
     if fixed_output_len is not None and fixed_output_len < 4:
         raise ValueError("output_len too small")
 
@@ -40,7 +35,7 @@ def sample_requests(
     random.shuffle(dataset)
 
     # Filter out sequences that are too long or too short
-    filtered_dataset: list[tuple[str, int, int]] = []
+    filtered_dataset: List[Tuple[str, int, int]] = []
     for i in range(len(dataset)):
         if len(filtered_dataset) == num_requests:
             break
@@ -60,7 +55,8 @@ def sample_requests(
             # Prune too long sequences.
             continue
 
-        priority = get_random_flag()
+        #Select a equi-probable random priority
+        priority = 0 if random.random() < 0.5 else 1
 
         filtered_dataset.append((prompt, prompt_len, output_len, priority))
 
@@ -68,20 +64,13 @@ def sample_requests(
 
 
 def run_vllm(
-    requests: list[tuple[str, int, int]],
+    requests: List[Tuple[str, int, int]],
     n: int,
     engine_args: EngineArgs,
-    disable_detokenize: bool = False,
 ) -> float:
     from vllm import LLM, SamplingParams
     llm = LLM(**dataclasses.asdict(engine_args))
 
-    assert all(
-        llm.llm_engine.model_config.max_model_len >= (request[1] + request[2])
-        for request in requests), (
-            "Please ensure that max_model_len is greater than the sum of"
-            " input_len and output_len for all requests.")
-
     # Add the requests to the engine.
     prompts = []
     sampling_params = []
@@ -96,7 +85,6 @@ def run_vllm(
                 top_p=1.0,
                 ignore_eos=True,
                 max_tokens=output_len,
-                detokenize=not disable_detokenize,
             ))
 
     start = time.perf_counter()
@@ -115,16 +103,15 @@ def main(args: argparse.Namespace):
     if args.dataset is None:
         # Synthesize a prompt with the given input length.
         prompt = "hi" * (args.input_len - 1)
-        requests = [(prompt, args.input_len, args.output_len,
-                     get_random_flag()) for _ in range(args.num_prompts)]
+        requests = [(prompt, args.input_len, args.output_len)
+                    for _ in range(args.num_prompts)]
     else:
         requests = sample_requests(args.dataset, args.num_prompts, tokenizer,
                                    args.output_len)
 
     if args.backend == "vllm":
         elapsed_time = run_vllm(requests, args.n,
-                                EngineArgs.from_cli_args(args),
-                                args.disable_detokenize)
+                                EngineArgs.from_cli_args(args))
     else:
         raise ValueError(f"Unknown backend: {args.backend}")
     total_num_tokens = sum(prompt_len + output_len
@@ -177,12 +164,6 @@ if __name__ == "__main__":
         type=str,
         default=None,
         help='Path to save the throughput results in JSON format.')
-    parser.add_argument(
-        '--disable-detokenize',
-        action='store_true',
-        help=("Do not detokenize responses (i.e. do not include "
-              "detokenization time in the latency measurement)"),
-    )
 
     parser = EngineArgs.add_cli_args(parser)
     args = parser.parse_args()

benchmarks/benchmark_serving_guided.py

@@ -1,17 +1,20 @@
 # SPDX-License-Identifier: Apache-2.0
-r"""Benchmark online serving throughput with structured outputs.
+r"""Benchmark online serving throughput with guided decoding.
 
 On the server side, run one of the following commands:
     (vLLM OpenAI API server)
     vllm serve <your_model> --disable-log-requests
 
+    (TGI backend)
+    ./launch_tgi_server.sh <your_model> <max_batch_total_tokens>
+
 On the client side, run:
-    python benchmarks/benchmark_serving_structured_output.py \
+    python benchmarks/benchmark_serving.py \
         --backend <backend> \
         --model <your_model> \
         --dataset json \
-        --structured-output-ratio 1.0 \
-        --structured-output-backend auto \
+        --guided-decoding-ratio 1.0 \
+        --guided-decoding-backend xgrammar \
         --request-rate 10 \
         --num-prompts 1000
 
@@ -21,17 +24,14 @@ On the client side, run:
 """
 import argparse
 import asyncio
-import copy
 import dataclasses
 import json
 import os
 import random
 import time
-import uuid
 import warnings
-from collections.abc import AsyncGenerator
 from dataclasses import dataclass
-from typing import Optional
+from typing import AsyncGenerator, List, Optional, Tuple
 
 import datasets
 import numpy as np
@@ -51,9 +51,6 @@ try:
 except ImportError:
     from argparse import ArgumentParser as FlexibleArgumentParser
 
-from vllm.v1.structured_output.backend_xgrammar import (
-    has_xgrammar_unsupported_json_features)
-
 MILLISECONDS_TO_SECONDS_CONVERSION = 1000
 
 
@@ -69,22 +66,22 @@ class BenchmarkMetrics:
     mean_ttft_ms: float
     median_ttft_ms: float
     std_ttft_ms: float
-    percentiles_ttft_ms: list[tuple[float, float]]
+    percentiles_ttft_ms: List[Tuple[float, float]]
     mean_tpot_ms: float
     median_tpot_ms: float
     std_tpot_ms: float
-    percentiles_tpot_ms: list[tuple[float, float]]
+    percentiles_tpot_ms: List[Tuple[float, float]]
     mean_itl_ms: float
     median_itl_ms: float
     std_itl_ms: float
-    percentiles_itl_ms: list[tuple[float, float]]
+    percentiles_itl_ms: List[Tuple[float, float]]
     # E2EL stands for end-to-end latency per request.
     # It is the time taken on the client side from sending
     # a request to receiving a complete response.
     mean_e2el_ms: float
     median_e2el_ms: float
     std_e2el_ms: float
-    percentiles_e2el_ms: list[tuple[float, float]]
+    percentiles_e2el_ms: List[Tuple[float, float]]
 
 
 @dataclasses.dataclass
@@ -107,62 +104,40 @@ class SampleRequest:
 
 
 def sample_requests(tokenizer: PreTrainedTokenizerBase,
-                    args: argparse.Namespace) -> list[SampleRequest]:
-    if args.dataset == 'json' or args.dataset == 'json-unique':
+                    args: argparse.Namespace) -> List[SampleRequest]:
+    if args.dataset == 'json':
         if args.json_schema_path is None:
             dir_path = os.path.dirname(os.path.realpath(__file__))
             args.json_schema_path = os.path.join(dir_path,
                                                  "structured_schemas",
                                                  "structured_schema_1.json")
-        json_schemas = []
         with open(args.json_schema_path) as f:
             schema = json.load(f)
-
-        if args.dataset == 'json-unique':
-            json_schemas = [
-                copy.deepcopy(schema) for _ in range(args.num_prompts)
-            ]
-            for i in range(len(json_schemas)):
-                if "properties" not in json_schemas[i]:
-                    json_schemas[i]["properties"] = {}
-                json_schemas[i]["properties"][
-                    f"__optional_field_{uuid.uuid4()}"] = {
-                        "type":
-                        "string",
-                        "description":
-                        "An unique optional field to avoid cached schemas"
-                    }
-        else:
-            json_schemas = [schema] * args.num_prompts
-
-        def gen_prompt(index: int):
-            return f"Generate an example of a brief user profile given the following schema: {json.dumps(get_schema(index))}"  # noqa: E501
-
-        def get_schema(index: int):
-            return json_schemas[index % len(json_schemas)]
-
+        prompt = f"Generate an example of a user profile given the following schema: {json.dumps(schema)}"  # noqa: E501
+        input_len = len(tokenizer(prompt).input_ids)
+        print(f"Input length of the prompt: {input_len} tokens")
         requests = [
-            SampleRequest(prompt=gen_prompt(i),
-                          prompt_len=len(tokenizer(gen_prompt(i)).input_ids),
+            SampleRequest(prompt=prompt,
+                          prompt_len=input_len,
                           expected_output_len=args.output_len,
-                          schema=get_schema(i),
+                          schema=schema,
                           structure_type=args.structure_type)
-            for i in range(args.num_prompts)
+            for _ in range(args.num_prompts)
         ]
 
     elif args.dataset == "grammar":
         schema = """
-        root ::= select_statement
+            ?start: select_statement
 
-        select_statement ::= "SELECT " column " from " table " where " condition
+            ?select_statement: "SELECT " column_list " FROM " table_name
 
-        column ::= "col_1 " | "col_2 "
+            ?column_list: column_name ("," column_name)*
 
-        table ::= "table_1 " | "table_2 "
+            ?table_name: identifier
 
-        condition ::= column "= " number
+            ?column_name: identifier
 
-        number ::= "1 " | "2 "
+            ?identifier: /[a-zA-Z_][a-zA-Z0-9_]*/
         """
         prompt = "Generate an SQL query to show the 'username' \
             and 'email' from the 'users' table."
@@ -212,20 +187,10 @@ def sample_requests(tokenizer: PreTrainedTokenizerBase,
         ]
 
     elif args.dataset == "xgrammar_bench":
-        requests: list[SampleRequest] = []
+        requests: List[SampleRequest] = []
         dataset = datasets.load_dataset("NousResearch/json-mode-eval",
                                         split="train")
-        full_dataset_len = len(dataset)
-
-        def _filter_func(item):
-            import json
-            schema = json.loads(item["schema"])
-            return not has_xgrammar_unsupported_json_features(schema)
-
-        dataset = dataset.filter(_filter_func)
-        num_filtered_out = full_dataset_len - len(dataset)
-        print(f"dataset has {len(dataset)} entries after filtering "
-              f"out {num_filtered_out} entries with unsupported features")
+        print(f"dataset has {len(dataset)} entries")
         len_dataset = len(dataset)
         for data_point_idx in range(args.num_prompts):
             idx = data_point_idx
@@ -233,8 +198,7 @@ def sample_requests(tokenizer: PreTrainedTokenizerBase,
                 idx -= len_dataset
             schema = dataset["schema"][idx]
             prompt = tokenizer.apply_chat_template(dataset["prompt"][idx],
-                                                   tokenize=False,
-                                                   add_generation_prompt=True)
+                                                   tokenize=False)
             input_len = len(tokenizer(prompt).input_ids)
             completion = dataset["completion"][idx]
 
@@ -250,26 +214,26 @@ def sample_requests(tokenizer: PreTrainedTokenizerBase,
 
 
 async def get_request(
-    input_requests: list[SampleRequest],
+    input_requests: List[SampleRequest],
     request_rate: float,
     burstiness: float = 1.0,
-) -> AsyncGenerator[tuple[int, SampleRequest], None]:
+) -> AsyncGenerator[Tuple[int, SampleRequest], None]:
     """
-    Asynchronously generates requests at a specified rate
+    Asynchronously generates requests at a specified rate 
     with OPTIONAL burstiness.
-
+    
     Args:
-        input_requests:
+        input_requests: 
             A list of input requests, each represented as a tuple.
-        request_rate:
+        request_rate: 
             The rate at which requests are generated (requests/s).
-        burstiness (optional):
-            The burstiness factor of the request generation.
+        burstiness (optional): 
+            The burstiness factor of the request generation. 
             Only takes effect when request_rate is not inf.
             Default value is 1, which follows a Poisson process.
             Otherwise, the request intervals follow a gamma distribution.
-            A lower burstiness value (0 < burstiness < 1) results
-            in more bursty requests, while a higher burstiness value
+            A lower burstiness value (0 < burstiness < 1) results 
+            in more bursty requests, while a higher burstiness value 
             (burstiness > 1) results in a more uniform arrival of requests.
     """
     input_requests = iter(input_requests)
@@ -294,23 +258,22 @@ async def get_request(
 
 
 def calculate_metrics(
-    input_requests: list[tuple[str, int, int]],
-    outputs: list[RequestFuncOutput],
+    input_requests: List[Tuple[str, int, int]],
+    outputs: List[RequestFuncOutput],
     dur_s: float,
     tokenizer: PreTrainedTokenizerBase,
-    selected_percentile_metrics: list[str],
-    selected_percentiles: list[float],
-    goodput_config_dict: Optional[dict[str, float]] = None,
-) -> tuple[BenchmarkMetrics, list[int]]:
-    actual_output_lens: list[int] = []
+    selected_percentile_metrics: List[str],
+    selected_percentiles: List[float],
+) -> Tuple[BenchmarkMetrics, List[int]]:
+    actual_output_lens: List[int] = []
     total_input = 0
     completed = 0
     good_completed = 0
-    itls: list[float] = []
-    tpots: list[float] = []
-    all_tpots: list[float] = []
-    ttfts: list[float] = []
-    e2els: list[float] = []
+    itls: List[float] = []
+    tpots: List[float] = []
+    all_tpots: List[float] = []
+    ttfts: List[float] = []
+    e2els: List[float] = []
     for i in range(len(outputs)):
         if outputs[i].success:
             # We use the tokenizer to count the number of output tokens for all
@@ -324,10 +287,10 @@ def calculate_metrics(
             total_input += input_requests[i].prompt_len
             tpot = 0
             if output_len > 1:
-                latency_minus_ttft = outputs[i].latency - outputs[i].ttft
-                tpot = latency_minus_ttft / (output_len - 1)
+                tpot = (outputs[i].latency - outputs[i].ttft) / (output_len -
+                                                                 1)
                 tpots.append(tpot)
-            outputs[i].tpot = tpot
+            outputs[i].tpot = sum(tpots) / len(tpots) if len(tpots) else 0
             # Note: if output_len <= 1, we regard tpot as 0 for goodput
             all_tpots.append(tpot)
             itls += outputs[i].itl
@@ -337,28 +300,6 @@ def calculate_metrics(
         else:
             actual_output_lens.append(0)
 
-    if goodput_config_dict:
-        valid_metrics = []
-        slo_values = []
-
-        if "ttft" in goodput_config_dict:
-            valid_metrics.append(ttfts)
-            slo_values.append(goodput_config_dict["ttft"] /
-                              MILLISECONDS_TO_SECONDS_CONVERSION)
-        if "tpot" in goodput_config_dict:
-            valid_metrics.append(all_tpots)
-            slo_values.append(goodput_config_dict["tpot"] /
-                              MILLISECONDS_TO_SECONDS_CONVERSION)
-        if "e2el" in goodput_config_dict:
-            valid_metrics.append(e2els)
-            slo_values.append(goodput_config_dict["e2el"] /
-                              MILLISECONDS_TO_SECONDS_CONVERSION)
-
-        for req_metric in zip(*valid_metrics):
-            is_good_req = all([s >= r for s, r in zip(slo_values, req_metric)])
-            if is_good_req:
-                good_completed += 1
-
     if completed == 0:
         warnings.warn(
             "All requests failed. This is likely due to a misconfiguration "
@@ -404,17 +345,17 @@ async def benchmark(
     base_url: str,
     model_id: str,
     tokenizer: PreTrainedTokenizerBase,
-    input_requests: list[SampleRequest],
+    input_requests: List[SampleRequest],
     request_rate: float,
     burstiness: float,
     disable_tqdm: bool,
     profile: bool,
-    selected_percentile_metrics: list[str],
-    selected_percentiles: list[str],
+    selected_percentile_metrics: List[str],
+    selected_percentiles: List[str],
     ignore_eos: bool,
     max_concurrency: Optional[int],
-    structured_output_ratio: float,
-    goodput_config_dict: Optional[dict[str, float]] = None,
+    guided_decoding_ratio: float,
+    guided_decoding_backend: str,
 ):
     if backend in ASYNC_REQUEST_FUNCS:
         request_func = ASYNC_REQUEST_FUNCS[backend]
@@ -425,16 +366,16 @@ async def benchmark(
         extra_body = {}
         # Add the schema to the extra_body
         extra_body[request.structure_type] = request.schema
+        # Add the specific guided_decoding_backend
+        extra_body["guided_decoding_backend"] = guided_decoding_backend
         return extra_body
 
     print("Starting initial single prompt test run...")
-    structured_output_req_idx = random.sample(
+    guided_decoding_req_idx = random.sample(
         range(len(input_requests)),
-        int(len(input_requests) * structured_output_ratio))
+        int(len(input_requests) * guided_decoding_ratio))
 
     test_request = input_requests[0]
-    test_req_extra_body = (prepare_extra_body(test_request)
-                           if 0 in structured_output_req_idx else None)
     test_input = RequestFuncInput(
         model=model_id,
         prompt=test_request.prompt,
@@ -442,7 +383,7 @@ async def benchmark(
         prompt_len=test_request.prompt_len,
         output_len=test_request.expected_output_len,
         ignore_eos=ignore_eos,
-        extra_body=test_req_extra_body,
+        extra_body=prepare_extra_body(test_request),
     )
     test_output = await request_func(request_func_input=test_input)
     if not test_output.success:
@@ -461,7 +402,7 @@ async def benchmark(
             prompt_len=test_request.prompt_len,
             output_len=test_request.expected_output_len,
             ignore_eos=ignore_eos,
-            extra_body=test_req_extra_body,
+            extra_body=prepare_extra_body(test_request),
         )
         profile_output = await request_func(request_func_input=profile_input)
         if profile_output.success:
@@ -494,12 +435,12 @@ async def benchmark(
                                       pbar=pbar)
 
     benchmark_start_time = time.perf_counter()
-    tasks: list[asyncio.Task] = []
-    expected: list[str] = []
+    tasks: List[asyncio.Task] = []
+    expected: List[str] = []
     async for i, request in get_request(input_requests, request_rate,
                                         burstiness):
         extra_body = prepare_extra_body(
-            request) if i in structured_output_req_idx else None
+            request) if i in guided_decoding_req_idx else None
         request_func_input = RequestFuncInput(
             model=model_id,
             prompt=request.prompt,
@@ -514,7 +455,7 @@ async def benchmark(
             asyncio.create_task(
                 limited_request_func(request_func_input=request_func_input,
                                      pbar=pbar)))
-    outputs: list[RequestFuncOutput] = await asyncio.gather(*tasks)
+    outputs: List[RequestFuncOutput] = await asyncio.gather(*tasks)
 
     if profile:
         print("Stopping profiler...")
@@ -542,7 +483,6 @@ async def benchmark(
         tokenizer=tokenizer,
         selected_percentile_metrics=selected_percentile_metrics,
         selected_percentiles=selected_percentiles,
-        goodput_config_dict=goodput_config_dict,
     )
 
     print("{s:{c}^{n}}".format(s=' Serving Benchmark Result ', n=50, c='='))
@@ -554,9 +494,6 @@ async def benchmark(
                                  metrics.total_output))
     print("{:<40} {:<10.2f}".format("Request throughput (req/s):",
                                     metrics.request_throughput))
-    if goodput_config_dict:
-        print("{:<40} {:<10.2f}".format("Request goodput (req/s):",
-                                        metrics.request_goodput))
     print("{:<40} {:<10.2f}".format("Output token throughput (tok/s):",
                                     metrics.output_throughput))
     print("{:<40} {:<10.2f}".format("Total Token throughput (tok/s):",
@@ -680,40 +617,6 @@ def evaluate(ret, args):
             100) if len(not_none_scores) > 0 else None
 
 
-def parse_goodput(slo_pairs):
-    goodput_config_dict = {}
-    try:
-        for slo_pair in slo_pairs:
-            slo_name, slo_val = slo_pair.split(":")
-            goodput_config_dict[slo_name] = float(slo_val)
-    except ValueError as err:
-        raise argparse.ArgumentTypeError(
-            "Invalid format found for service level objectives. "
-            "Specify service level objectives for goodput as \"KEY:VALUE\" "
-            "pairs, where the key is a metric name, and the value is a "
-            "number in milliseconds.") from err
-    return goodput_config_dict
-
-
-def check_goodput_args(args):
-    goodput_config_dict = {}
-    VALID_NAMES = ["ttft", "tpot", "e2el"]
-    if args.goodput:
-        goodput_config_dict = parse_goodput(args.goodput)
-        for slo_name, slo_val in goodput_config_dict.items():
-            if slo_name not in VALID_NAMES:
-                raise ValueError(
-                    f"Invalid metric name found, {slo_name}: {slo_val}. "
-                    "The service level objective name should be one of "
-                    f"{str(VALID_NAMES)}. ")
-            if slo_val < 0:
-                raise ValueError(
-                    f"Invalid value found, {slo_name}: {slo_val}. "
-                    "The service level objective value should be "
-                    "non-negative.")
-    return goodput_config_dict
-
-
 def main(args: argparse.Namespace):
     print(args)
     random.seed(args.seed)
@@ -730,11 +633,8 @@ def main(args: argparse.Namespace):
         api_url = f"http://{args.host}:{args.port}{args.endpoint}"
         base_url = f"http://{args.host}:{args.port}"
 
-    tokenizer = get_tokenizer(
-        tokenizer_id,
-        trust_remote_code=args.trust_remote_code,
-        tokenizer_mode=args.tokenizer_mode,
-    )
+    tokenizer = get_tokenizer(tokenizer_id,
+                              trust_remote_code=args.trust_remote_code)
 
     if args.dataset == 'grammar':
         args.structure_type = 'guided_grammar'
@@ -745,10 +645,10 @@ def main(args: argparse.Namespace):
     else:
         args.structure_type = 'guided_json'
 
-    if args.no_structured_output:
-        args.structured_output_ratio = 0
+    if args.no_guided_decoding:
+        args.guided_decoding_ratio = 0
     if args.save_results:
-        result_file_name = f'{args.structured_output_ratio}guided'
+        result_file_name = f'{args.guided_decoding_ratio}guided'
         result_file_name += f"_{backend}"
         result_file_name += f"_{args.request_rate}qps"
         result_file_name += f"_{args.model.split('/')[-1]}"
@@ -761,8 +661,6 @@ def main(args: argparse.Namespace):
 
     input_requests = sample_requests(tokenizer, args)
 
-    goodput_config_dict = check_goodput_args(args)
-
     benchmark_result, ret = asyncio.run(
         benchmark(
             backend=backend,
@@ -781,8 +679,8 @@ def main(args: argparse.Namespace):
             ],
             ignore_eos=args.ignore_eos,
             max_concurrency=args.max_concurrency,
-            structured_output_ratio=args.structured_output_ratio,
-            goodput_config_dict=goodput_config_dict,
+            guided_decoding_ratio=args.guided_decoding_ratio,
+            guided_decoding_backend=args.guided_decoding_backend,
         ))
 
     # Save config and results to json
@@ -833,8 +731,7 @@ if __name__ == "__main__":
         default=None,
         help="Server or API base url if not using http host and port.",
     )
-    # Use 127.0.0.1 here instead of localhost to force the use of ipv4
-    parser.add_argument("--host", type=str, default="127.0.0.1")
+    parser.add_argument("--host", type=str, default="localhost")
     parser.add_argument("--port", type=int, default=8000)
     parser.add_argument(
         "--endpoint",
@@ -842,13 +739,11 @@ if __name__ == "__main__":
         default="/v1/completions",
         help="API endpoint.",
     )
-    parser.add_argument("--dataset",
-                        default='json',
-                        choices=[
-                            'json', 'json-unique', 'grammar', 'regex',
-                            'choice', 'xgrammar_bench'
-                        ])
-    parser.add_argument("--json-schema-path",
+    parser.add_argument(
+        "--dataset",
+        default='json',
+        choices=['json', 'grammar', 'regex', 'choice', 'xgrammar_bench'])
+    parser.add_argument("--json_schema_path",
                         type=str,
                         default=None,
                         help="Path to json schema.")
@@ -876,13 +771,6 @@ if __name__ == "__main__":
         help=
         "Name or path of the tokenizer, if not using the default tokenizer.",  # noqa: E501
     )
-    parser.add_argument(
-        "--tokenizer-mode",
-        type=str,
-        default="auto",
-        help=
-        "Name or path of the tokenizer, if not using the default tokenizer.",  # noqa: E501
-    )
     parser.add_argument(
         "--num-prompts",
         type=int,
@@ -963,7 +851,7 @@ if __name__ == "__main__":
         "--percentile-metrics",
         type=str,
         default="ttft,tpot,itl",
-        help="Comma-separated list of selected metrics to report percentils. "
+        help="Comma-seperated list of selected metrics to report percentils. "
         "This argument specifies the metrics to report percentiles. "
         "Allowed metric names are \"ttft\", \"tpot\", \"itl\", \"e2el\". "
         "Default value is \"ttft,tpot,itl\".")
@@ -971,31 +859,24 @@ if __name__ == "__main__":
         "--metric-percentiles",
         type=str,
         default="99",
-        help="Comma-separated list of percentiles for selected metrics. "
+        help="Comma-seperated list of percentiles for selected metrics. "
         "To report 25-th, 50-th, and 75-th percentiles, use \"25,50,75\". "
         "Default value is \"99\". "
         "Use \"--percentile-metrics\" to select metrics.",
     )
-    parser.add_argument(
-        "--goodput",
-        nargs="+",
-        required=False,
-        help="Specify service level objectives for goodput as \"KEY:VALUE\" "
-        "pairs, where the key is a metric name, and the value is in "
-        "milliseconds. Multiple \"KEY:VALUE\" pairs can be provided, "
-        "separated by spaces. Allowed request level metric names are "
-        "\"ttft\", \"tpot\", \"e2el\". For more context on the definition of "
-        "goodput, refer to DistServe paper: https://arxiv.org/pdf/2401.09670 "
-        "and the blog: https://hao-ai-lab.github.io/blogs/distserve")
-
-    parser.add_argument("--no-structured-output",
+    parser.add_argument("--no-guided-decoding",
                         action='store_true',
                         default=False,
                         help="Whether to disable JSON decoding or not.")
-    parser.add_argument("--structured-output-ratio",
+    parser.add_argument("--guided-decoding-ratio",
                         type=float,
                         default=1.0,
-                        help="Ratio of Structured Outputs requests")
+                        help="Ratio of Guided Decoding requests")
+    parser.add_argument("--guided-decoding-backend",
+                        type=str,
+                        choices=["outlines", "lm-format-enforcer", "xgrammar"],
+                        default="xgrammar",
+                        help="Backend to use for guided decoding")
 
     args = parser.parse_args()
     main(args)

benchmarks/benchmark_throughput.py

@@ -3,19 +3,14 @@
 import argparse
 import dataclasses
 import json
-import os
 import random
 import time
-import warnings
-from typing import Any, Optional, Union
+from functools import cache
+from typing import Dict, List, Optional, Tuple
 
 import torch
 import uvloop
-from benchmark_dataset import (AIMODataset, BurstGPTDataset,
-                               ConversationDataset, InstructCoderDataset,
-                               RandomDataset, SampleRequest, ShareGPTDataset,
-                               SonnetDataset, VisionArenaDataset)
-from benchmark_utils import convert_to_pytorch_benchmark_format, write_to_json
+from PIL import Image
 from tqdm import tqdm
 from transformers import (AutoModelForCausalLM, AutoTokenizer,
                           PreTrainedTokenizerBase)
@@ -23,35 +18,163 @@ from transformers import (AutoModelForCausalLM, AutoTokenizer,
 from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs
 from vllm.entrypoints.openai.api_server import (
     build_async_engine_client_from_engine_args)
-from vllm.inputs import TextPrompt, TokensPrompt
+from vllm.inputs import TextPrompt
 from vllm.lora.request import LoRARequest
-from vllm.outputs import RequestOutput
+from vllm.lora.utils import get_adapter_absolute_path
+from vllm.multimodal import MultiModalDataDict
 from vllm.sampling_params import BeamSearchParams
+from vllm.transformers_utils.tokenizer import AnyTokenizer, get_lora_tokenizer
 from vllm.utils import FlexibleArgumentParser, merge_async_iterators
 
 
+@dataclasses.dataclass
+class SampleRequest:
+    """A class representing a single inference request for benchmarking.
+
+    Attributes:
+        prompt: The input text prompt for the model.
+        prompt_len: The length of the prompt in tokens.
+        expected_output_len: The expected length of the output in tokens.
+        multi_modal_data: Optional dictionary containing multi-modal data (e.g.
+            images).
+        lora_request: Optional LoRARequest specifying the LoRA to use. 
+    """
+    prompt: str
+    prompt_len: int
+    expected_output_len: int
+    multi_modal_data: Optional[MultiModalDataDict] = None
+    lora_request: Optional[LoRARequest] = None
+
+
+def _get_prompt_for_image_model(question: str, *, model: str) -> str:
+    """Prepend and append special tokens around the question to form a prompt.
+
+    Args:
+        question: The input question text to wrap with special tokens
+        model: The name of the model being used, to determine which special
+            tokens to add
+
+    Returns:
+        The formatted prompt string with appropriate special tokens for the
+            model
+
+    Raises:
+        ValueError: If an unsupported model name is provided
+    """
+    model = model.lower()
+    if "pixtral" in model:
+        return f"<s>[INST]{question}\n[IMG][/INST]"
+    raise ValueError(f"Unsupported model {model}")
+
+
+@cache
+def lora_path_on_disk(lora_path: str) -> str:
+    return get_adapter_absolute_path(lora_path)
+
+
+lora_tokenizer_cache: Dict[int, AnyTokenizer] = {}
+
+
+def get_random_lora_request(
+        args: argparse.Namespace
+) -> Tuple[LoRARequest, Optional[AnyTokenizer]]:
+    global lora_tokenizer_cache
+    lora_id = random.randint(1, args.max_loras)
+    lora_request = LoRARequest(lora_name=str(lora_id),
+                               lora_int_id=lora_id,
+                               lora_path=lora_path_on_disk(args.lora_path))
+    if lora_id not in lora_tokenizer_cache:
+        lora_tokenizer_cache[lora_id] = get_lora_tokenizer(lora_request)
+    return lora_request, lora_tokenizer_cache[lora_id]
+
+
+def sample_requests(tokenizer: PreTrainedTokenizerBase,
+                    args: argparse.Namespace) -> List[SampleRequest]:
+
+    dataset_path: str = args.dataset
+    num_requests: int = args.num_prompts
+    fixed_output_len: Optional[int] = args.output_len
+    model: str = args.model
+    if fixed_output_len is not None and fixed_output_len < 4:
+        raise ValueError("output_len too small")
+
+    # Load the dataset.
+    with open(dataset_path) as f:
+        dataset = json.load(f)
+    # Filter out the conversations with less than 2 turns.
+    dataset = [data for data in dataset if len(data["conversations"]) >= 2]
+    # Shuffle the dataset.
+    random.shuffle(dataset)
+
+    # Filter out sequences that are too long or too short
+    filtered_dataset: List[SampleRequest] = []
+    for data in tqdm(dataset,
+                     total=len(filtered_dataset),
+                     desc="sampling requests"):
+        if len(filtered_dataset) == num_requests:
+            break
+
+        # Only keep the first two turns of each conversation.
+        prompt = data["conversations"][0]["value"]
+        completion = data["conversations"][1]["value"]
+
+        multi_modal_data: Optional[MultiModalDataDict] = None
+        if "image" in data:
+            multi_modal_data = multi_modal_data or {}
+            image_path = data["image"]
+            # TODO(vllm-project/vllm/issues/9778): Support multiple images.
+            assert isinstance(image_path,
+                              str), "Only support single image input"
+            try:
+                multi_modal_data["image"] = Image.open(image_path).convert(
+                    "RGB")
+            except FileNotFoundError:
+                # Ignore datapoint where asset is missing
+                continue
+            prompt = _get_prompt_for_image_model(question=prompt, model=model)
+
+        request_tokenizer = tokenizer
+        lora_request: Optional[LoRARequest] = None
+        if args.enable_lora:
+            lora_request, lora_tokenizer = get_random_lora_request(args)
+            if lora_tokenizer:
+                request_tokenizer = lora_tokenizer
+
+        # Tokenize the prompts and completions.
+        prompt_token_ids = request_tokenizer(prompt).input_ids
+        completion_token_ids = request_tokenizer(completion).input_ids
+        prompt_len = len(prompt_token_ids)
+        output_len = len(completion_token_ids
+                         ) if fixed_output_len is None else fixed_output_len
+        if prompt_len < 4 or output_len < 4:
+            # Prune too short sequences.
+            continue
+        if prompt_len > 1024 or prompt_len + output_len > 2048:
+            # Prune too long sequences.
+            continue
+        filtered_dataset.append(
+            SampleRequest(prompt=prompt,
+                          prompt_len=prompt_len,
+                          expected_output_len=output_len,
+                          multi_modal_data=multi_modal_data,
+                          lora_request=lora_request))
+
+    return filtered_dataset
+
+
 def run_vllm(
-    requests: list[SampleRequest],
+    requests: List[SampleRequest],
     n: int,
     engine_args: EngineArgs,
-    disable_detokenize: bool = False,
-) -> tuple[float, Optional[list[RequestOutput]]]:
+) -> float:
     from vllm import LLM, SamplingParams
     llm = LLM(**dataclasses.asdict(engine_args))
-    assert all(
-        llm.llm_engine.model_config.max_model_len >= (
-            request.prompt_len + request.expected_output_len)
-        for request in requests), (
-            "Please ensure that max_model_len is greater than the sum of"
-            " prompt_len and expected_output_len for all requests.")
+
     # Add the requests to the engine.
-    prompts: list[Union[TextPrompt, TokensPrompt]] = []
-    sampling_params: list[SamplingParams] = []
+    prompts: List[TextPrompt] = []
+    sampling_params: List[SamplingParams] = []
     for request in requests:
         prompts.append(
-            TokensPrompt(prompt_token_ids=request.prompt["prompt_token_ids"],
-                       multi_modal_data=request.multi_modal_data)
-            if "prompt_token_ids" in request.prompt else \
             TextPrompt(prompt=request.prompt,
                        multi_modal_data=request.multi_modal_data))
         sampling_params.append(
@@ -61,21 +184,19 @@ def run_vllm(
                 top_p=1.0,
                 ignore_eos=True,
                 max_tokens=request.expected_output_len,
-                detokenize=not disable_detokenize,
             ))
-    lora_requests: Optional[list[LoRARequest]] = None
+    lora_requests: Optional[List[LoRARequest]] = None
     if engine_args.enable_lora:
         lora_requests = [request.lora_request for request in requests]
 
     use_beam_search = False
 
-    outputs = None
     if not use_beam_search:
         start = time.perf_counter()
-        outputs = llm.generate(prompts,
-                               sampling_params,
-                               lora_request=lora_requests,
-                               use_tqdm=True)
+        llm.generate(prompts,
+                     sampling_params,
+                     lora_request=lora_requests,
+                     use_tqdm=True)
         end = time.perf_counter()
     else:
         assert lora_requests is None, "BeamSearch API does not support LoRA"
@@ -93,75 +214,26 @@ def run_vllm(
                 ignore_eos=True,
             ))
         end = time.perf_counter()
-    return end - start, outputs
-
-
-def run_vllm_chat(
-        requests: list[SampleRequest],
-        n: int,
-        engine_args: EngineArgs,
-        disable_detokenize: bool = False) -> tuple[float, list[RequestOutput]]:
-    """
-    Run vLLM chat benchmark. This function is recommended ONLY for benchmarking
-    multimodal models as it properly handles multimodal inputs and chat
-    formatting. For non-multimodal models, use run_vllm() instead.
-    """
-    from vllm import LLM, SamplingParams
-    llm = LLM(**dataclasses.asdict(engine_args))
-
-    assert all(
-        llm.llm_engine.model_config.max_model_len >= (
-            request.prompt_len + request.expected_output_len)
-        for request in requests), (
-            "Please ensure that max_model_len is greater than the sum of "
-            "prompt_len and expected_output_len for all requests.")
-
-    prompts = []
-    sampling_params: list[SamplingParams] = []
-    for request in requests:
-        prompts.append(request.prompt)
-        sampling_params.append(
-            SamplingParams(
-                n=n,
-                temperature=1.0,
-                top_p=1.0,
-                ignore_eos=True,
-                max_tokens=request.expected_output_len,
-                detokenize=not disable_detokenize,
-            ))
-    start = time.perf_counter()
-    outputs = llm.chat(prompts, sampling_params, use_tqdm=True)
-    end = time.perf_counter()
-    return end - start, outputs
+    return end - start
 
 
 async def run_vllm_async(
-    requests: list[SampleRequest],
+    requests: List[SampleRequest],
     n: int,
     engine_args: AsyncEngineArgs,
     disable_frontend_multiprocessing: bool = False,
-    disable_detokenize: bool = False,
 ) -> float:
     from vllm import SamplingParams
 
     async with build_async_engine_client_from_engine_args(
             engine_args, disable_frontend_multiprocessing) as llm:
-        assert all(
-            llm.model_config.max_model_len >= (request.prompt_len +
-                                               request.expected_output_len)
-            for request in requests), (
-                "Please ensure that max_model_len is greater than the sum of"
-                " prompt_len and expected_output_len for all requests.")
 
         # Add the requests to the engine.
-        prompts: list[Union[TextPrompt, TokensPrompt]] = []
-        sampling_params: list[SamplingParams] = []
-        lora_requests: list[Optional[LoRARequest]] = []
+        prompts: List[TextPrompt] = []
+        sampling_params: List[SamplingParams] = []
+        lora_requests: List[Optional[LoRARequest]] = []
         for request in requests:
             prompts.append(
-                TokensPrompt(prompt_token_ids=request.prompt["prompt_token_ids"],
-                        multi_modal_data=request.multi_modal_data)
-                if "prompt_token_ids" in request.prompt else \
                 TextPrompt(prompt=request.prompt,
                            multi_modal_data=request.multi_modal_data))
             sampling_params.append(
@@ -171,7 +243,6 @@ async def run_vllm_async(
                     top_p=1.0,
                     ignore_eos=True,
                     max_tokens=request.expected_output_len,
-                    detokenize=not disable_detokenize,
                 ))
             lora_requests.append(request.lora_request)
 
@@ -192,13 +263,12 @@ async def run_vllm_async(
 
 
 def run_hf(
-    requests: list[SampleRequest],
+    requests: List[SampleRequest],
     model: str,
     tokenizer: PreTrainedTokenizerBase,
     n: int,
     max_batch_size: int,
     trust_remote_code: bool,
-    disable_detokenize: bool = False,
 ) -> float:
     llm = AutoModelForCausalLM.from_pretrained(
         model, torch_dtype=torch.float16, trust_remote_code=trust_remote_code)
@@ -209,21 +279,18 @@ def run_hf(
 
     pbar = tqdm(total=len(requests))
     start = time.perf_counter()
-    batch: list[str] = []
+    batch: List[str] = []
     max_prompt_len = 0
     max_output_len = 0
     for i in range(len(requests)):
-        prompt = requests[i].prompt
-        prompt_len = requests[i].prompt_len
-        output_len = requests[i].expected_output_len
+        prompt, prompt_len, output_len = requests[i]
         # Add the prompt to the batch.
         batch.append(prompt)
         max_prompt_len = max(max_prompt_len, prompt_len)
         max_output_len = max(max_output_len, output_len)
         if len(batch) < max_batch_size and i != len(requests) - 1:
             # Check if we can add more requests to the batch.
-            next_prompt_len = requests[i + 1].prompt_len
-            next_output_len = requests[i + 1].expected_output_len
+            _, next_prompt_len, next_output_len = requests[i + 1]
             if (max(max_prompt_len, next_prompt_len) +
                     max(max_output_len, next_output_len)) <= 2048:
                 # We can add more requests to the batch.
@@ -241,9 +308,8 @@ def run_hf(
             use_cache=True,
             max_new_tokens=max_output_len,
         )
-        if not disable_detokenize:
-            # Include the decoding time.
-            tokenizer.batch_decode(llm_outputs, skip_special_tokens=True)
+        # Include the decoding time.
+        tokenizer.batch_decode(llm_outputs, skip_special_tokens=True)
         pbar.update(len(batch))
 
         # Clear the batch.
@@ -255,7 +321,7 @@ def run_hf(
 
 
 def run_mii(
-    requests: list[SampleRequest],
+    requests: List[SampleRequest],
     model: str,
     tensor_parallel_size: int,
     output_len: int,
@@ -272,92 +338,58 @@ def run_mii(
     return end - start
 
 
-def save_to_pytorch_benchmark_format(args: argparse.Namespace,
-                                     results: dict[str, Any]) -> None:
-    pt_records = convert_to_pytorch_benchmark_format(
-        args=args,
-        metrics={
-            "requests_per_second": [results["requests_per_second"]],
-            "tokens_per_second": [results["tokens_per_second"]],
-        },
-        extra_info={
-            k: results[k]
-            for k in ["elapsed_time", "num_requests", "total_num_tokens"]
-        })
-    if pt_records:
-        # Don't use json suffix here as we don't want CI to pick it up
-        pt_file = f"{os.path.splitext(args.output_json)[0]}.pytorch.json"
-        write_to_json(pt_file, pt_records)
-
-
-def get_requests(args, tokenizer):
-    # Common parameters for all dataset types.
-    common_kwargs = {
-        "dataset_path": args.dataset_path,
-        "random_seed": args.seed,
-    }
-    sample_kwargs = {
-        "tokenizer": tokenizer,
-        "lora_path": args.lora_path,
-        "max_loras": args.max_loras,
-        "num_requests": args.num_prompts,
-        "input_len": args.input_len,
-        "output_len": args.output_len,
-    }
-
-    if args.dataset_path is None or args.dataset_name == "random":
-        sample_kwargs["range_ratio"] = args.random_range_ratio
-        sample_kwargs["prefix_len"] = args.prefix_len
-        dataset_cls = RandomDataset
-    elif args.dataset_name == "sharegpt":
-        dataset_cls = ShareGPTDataset
-        if args.backend == "vllm-chat":
-            sample_kwargs["enable_multimodal_chat"] = True
-    elif args.dataset_name == "sonnet":
-        assert tokenizer.chat_template or tokenizer.default_chat_template, (
-            "Tokenizer/model must have chat template for sonnet dataset.")
-        dataset_cls = SonnetDataset
-        sample_kwargs["prefix_len"] = args.prefix_len
-        sample_kwargs["return_prompt_formatted"] = True
-    elif args.dataset_name == "burstgpt":
-        dataset_cls = BurstGPTDataset
-    elif args.dataset_name == "hf":
-        if args.dataset_path in VisionArenaDataset.SUPPORTED_DATASET_PATHS:
-            dataset_cls = VisionArenaDataset
-            common_kwargs['dataset_subset'] = None
-            common_kwargs['dataset_split'] = "train"
-            sample_kwargs["enable_multimodal_chat"] = True
-        elif args.dataset_path in InstructCoderDataset.SUPPORTED_DATASET_PATHS:
-            dataset_cls = InstructCoderDataset
-            common_kwargs['dataset_split'] = "train"
-        elif args.dataset_path in ConversationDataset.SUPPORTED_DATASET_PATHS:
-            dataset_cls = ConversationDataset
-            common_kwargs['dataset_subset'] = args.hf_subset
-            common_kwargs['dataset_split'] = args.hf_split
-            sample_kwargs["enable_multimodal_chat"] = True
-        elif args.dataset_path in AIMODataset.SUPPORTED_DATASET_PATHS:
-            dataset_cls = AIMODataset
-            common_kwargs['dataset_subset'] = None
-            common_kwargs['dataset_split'] = "train"
-    else:
-        raise ValueError(f"Unknown dataset name: {args.dataset_name}")
-    # Remove None values
-    sample_kwargs = {k: v for k, v in sample_kwargs.items() if v is not None}
-    return dataset_cls(**common_kwargs).sample(**sample_kwargs)
-
-
 def main(args: argparse.Namespace):
-    if args.seed is None:
-        args.seed = 0
     print(args)
     random.seed(args.seed)
+
     # Sample the requests.
     tokenizer = AutoTokenizer.from_pretrained(
         args.tokenizer, trust_remote_code=args.trust_remote_code)
-    requests = get_requests(args, tokenizer)
+    if args.dataset is None:
+        vocab_size = tokenizer.vocab_size
+        requests = []
+        for _ in range(args.num_prompts):
+
+            request_tokenizer = tokenizer
+            lora_request: Optional[LoRARequest] = None
+            if args.enable_lora:
+                lora_request, lora_tokenizer = get_random_lora_request(args)
+                if lora_tokenizer:
+                    request_tokenizer = lora_tokenizer
+
+            # Synthesize a prompt with the given input length.
+            candidate_ids = [
+                random.randint(0, vocab_size - 1)
+                for _ in range(args.input_len)
+            ]
+            # As tokenizer may add additional tokens like BOS, we need to try
+            # different lengths to get the desired input length.
+            for _ in range(5):  # Max attempts to correct
+                candidate_prompt = request_tokenizer.decode(candidate_ids)
+                tokenized_len = len(request_tokenizer.encode(candidate_prompt))
+
+                if tokenized_len == args.input_len:
+                    break
+
+                # Adjust length based on difference
+                diff = args.input_len - tokenized_len
+                if diff > 0:
+                    candidate_ids.extend([
+                        random.randint(100, vocab_size - 100)
+                        for _ in range(diff)
+                    ])
+                else:
+                    candidate_ids = candidate_ids[:diff]
+            requests.append(
+                SampleRequest(prompt=candidate_prompt,
+                              prompt_len=args.input_len,
+                              expected_output_len=args.output_len,
+                              lora_request=lora_request))
+    else:
+        requests = sample_requests(tokenizer, args)
+
     is_multi_modal = any(request.multi_modal_data is not None
                          for request in requests)
-    request_outputs: Optional[list[RequestOutput]] = None
     if args.backend == "vllm":
         if args.async_engine:
             elapsed_time = uvloop.run(
@@ -366,59 +398,31 @@ def main(args: argparse.Namespace):
                     args.n,
                     AsyncEngineArgs.from_cli_args(args),
                     args.disable_frontend_multiprocessing,
-                    args.disable_detokenize,
                 ))
         else:
-            elapsed_time, request_outputs = run_vllm(
-                requests, args.n, EngineArgs.from_cli_args(args),
-                args.disable_detokenize)
+            elapsed_time = run_vllm(requests, args.n,
+                                    EngineArgs.from_cli_args(args))
     elif args.backend == "hf":
         assert args.tensor_parallel_size == 1
         elapsed_time = run_hf(requests, args.model, tokenizer, args.n,
-                              args.hf_max_batch_size, args.trust_remote_code,
-                              args.disable_detokenize)
+                              args.hf_max_batch_size, args.trust_remote_code)
     elif args.backend == "mii":
         elapsed_time = run_mii(requests, args.model, args.tensor_parallel_size,
                                args.output_len)
-    elif args.backend == "vllm-chat":
-        elapsed_time, request_outputs = run_vllm_chat(
-            requests, args.n, EngineArgs.from_cli_args(args),
-            args.disable_detokenize)
     else:
         raise ValueError(f"Unknown backend: {args.backend}")
-
-    if request_outputs:
-        # Note: with the vllm and vllm-chat backends,
-        # we have request_outputs, which we use to count tokens.
-        total_prompt_tokens = 0
-        total_output_tokens = 0
-        for ro in request_outputs:
-            if not isinstance(ro, RequestOutput):
-                continue
-            total_prompt_tokens += len(
-                ro.prompt_token_ids) if ro.prompt_token_ids else 0
-            total_output_tokens += sum(
-                len(o.token_ids) for o in ro.outputs if o)
-        total_num_tokens = total_prompt_tokens + total_output_tokens
-    else:
-        total_num_tokens = sum(r.prompt_len + r.expected_output_len
-                               for r in requests)
-        total_output_tokens = sum(r.expected_output_len for r in requests)
-        total_prompt_tokens = total_num_tokens - total_output_tokens
-
-    if is_multi_modal and args.backend != "vllm-chat":
-        print("\033[91mWARNING\033[0m: Multi-modal request with "
-              f"{args.backend} backend detected. The "
+    total_num_tokens = sum(request.prompt_len + request.expected_output_len
+                           for request in requests)
+    total_output_tokens = sum(request.expected_output_len
+                              for request in requests)
+    if is_multi_modal:
+        print("\033[91mWARNING\033[0m: Multi-modal request detected. The "
               "following metrics are not accurate because image tokens are not"
               " counted. See vllm-project/vllm/issues/9778 for details.")
-        # TODO(vllm-project/vllm/issues/9778): Count multi-modal token length.
-        # vllm-chat backend counts the image tokens now
-
+        # TODO(vllm-project/vllm/issues/9778): Count molti-modal token length.
     print(f"Throughput: {len(requests) / elapsed_time:.2f} requests/s, "
           f"{total_num_tokens / elapsed_time:.2f} total tokens/s, "
           f"{total_output_tokens / elapsed_time:.2f} output tokens/s")
-    print(f"Total num prompt tokens:  {total_prompt_tokens}")
-    print(f"Total num output tokens:  {total_output_tokens}")
 
     # Output JSON results if specified
     if args.output_json:
@@ -431,130 +435,20 @@ def main(args: argparse.Namespace):
         }
         with open(args.output_json, "w") as f:
             json.dump(results, f, indent=4)
-        save_to_pytorch_benchmark_format(args, results)
-
-
-def validate_args(args):
-    """
-    Validate command-line arguments.
-    """
-
-    # === Deprecation and Defaulting ===
-    if args.dataset is not None:
-        warnings.warn(
-            "The '--dataset' argument will be deprecated in the next release. "
-            "Please use '--dataset-name' and '--dataset-path' instead.",
-            stacklevel=2)
-        args.dataset_path = args.dataset
-
-    if not getattr(args, "tokenizer", None):
-        args.tokenizer = args.model
-
-    # === Backend Validation ===
-    valid_backends = {"vllm", "hf", "mii", "vllm-chat"}
-    if args.backend not in valid_backends:
-        raise ValueError(f"Unsupported backend: {args.backend}")
-
-    # === Dataset Configuration ===
-    if not args.dataset and not args.dataset_path:
-        print(
-            "When dataset path is not set, it will default to random dataset")
-        args.dataset_name = 'random'
-        if args.input_len is None:
-            raise ValueError("input_len must be provided for a random dataset")
-
-    # === Dataset Name Specific Checks ===
-    # --hf-subset and --hf-split: only used
-    # when dataset_name is 'hf'
-    if args.dataset_name != "hf" and (
-            getattr(args, "hf_subset", None) is not None
-            or getattr(args, "hf_split", None) is not None):
-        warnings.warn("--hf-subset and --hf-split will be ignored \
-                since --dataset-name is not 'hf'.",
-                      stacklevel=2)
-    elif args.dataset_name == "hf":
-        if args.dataset_path in (
-                VisionArenaDataset.SUPPORTED_DATASET_PATHS.keys()
-                | ConversationDataset.SUPPORTED_DATASET_PATHS):
-            assert args.backend == "vllm-chat", f"{args.dataset_path} needs to use vllm-chat as the backend."  #noqa: E501
-        elif args.dataset_path in (InstructCoderDataset.SUPPORTED_DATASET_PATHS
-                                   | AIMODataset.SUPPORTED_DATASET_PATHS):
-            assert args.backend == "vllm", f"{args.dataset_path} needs to use vllm as the backend."  #noqa: E501
-        else:
-            raise ValueError(
-                f"{args.dataset_path} is not supported by hf dataset.")
-
-    # --random-range-ratio: only used when dataset_name is 'random'
-    if args.dataset_name != 'random' and args.random_range_ratio is not None:
-        warnings.warn("--random-range-ratio will be ignored since \
-                --dataset-name is not 'random'.",
-                      stacklevel=2)
-
-    # --prefix-len: only used when dataset_name is 'random', 'sonnet', or not
-    # set.
-    if args.dataset_name not in {"random", "sonnet", None
-                                 } and args.prefix_len is not None:
-        warnings.warn("--prefix-len will be ignored since --dataset-name\
-                 is not 'random', 'sonnet', or not set.",
-                      stacklevel=2)
-
-    # === LoRA Settings ===
-    if getattr(args, "enable_lora", False) and args.backend != "vllm":
-        raise ValueError(
-            "LoRA benchmarking is only supported for vLLM backend")
-    if getattr(args, "enable_lora", False) and args.lora_path is None:
-        raise ValueError("LoRA path must be provided when enable_lora is True")
-
-    # === Backend-specific Validations ===
-    if args.backend == "hf" and args.hf_max_batch_size is None:
-        raise ValueError("HF max batch size is required for HF backend")
-    if args.backend != "hf" and args.hf_max_batch_size is not None:
-        raise ValueError("HF max batch size is only for HF backend.")
-
-    if args.backend in {"hf", "mii"} and getattr(args, "quantization",
-                                                 None) is not None:
-        raise ValueError("Quantization is only for vLLM backend.")
-
-    if args.backend == "mii" and args.dtype != "auto":
-        raise ValueError("dtype must be auto for MII backend.")
-    if args.backend == "mii" and args.n != 1:
-        raise ValueError("n must be 1 for MII backend.")
-    if args.backend == "mii" and args.tokenizer != args.model:
-        raise ValueError(
-            "Tokenizer must be the same as the model for MII backend.")
-
-    # --data-parallel is not supported currently.
-    # https://github.com/vllm-project/vllm/issues/16222
-    if args.data_parallel_size > 1:
-        raise ValueError(
-            "Data parallel is not supported in offline benchmark, \
-            please use benchmark serving instead")
 
 
 if __name__ == "__main__":
     parser = FlexibleArgumentParser(description="Benchmark the throughput.")
     parser.add_argument("--backend",
                         type=str,
-                        choices=["vllm", "hf", "mii", "vllm-chat"],
+                        choices=["vllm", "hf", "mii"],
                         default="vllm")
-    parser.add_argument(
-        "--dataset-name",
-        type=str,
-        choices=["sharegpt", "random", "sonnet", "burstgpt", "hf"],
-        help="Name of the dataset to benchmark on.",
-        default="sharegpt")
-    parser.add_argument(
-        "--dataset",
-        type=str,
-        default=None,
-        help="Path to the ShareGPT dataset, will be deprecated in\
-            the next release. The dataset is expected to "
-        "be a json in form of list[dict[..., conversations: "
-        "list[dict[..., value: <prompt_or_response>]]]]")
-    parser.add_argument("--dataset-path",
+    parser.add_argument("--dataset",
                         type=str,
                         default=None,
-                        help="Path to the dataset")
+                        help="Path to the dataset. The dataset is expected to "
+                        "be a json in form of List[Dict[..., conversations: "
+                        "List[Dict[..., value: <prompt_or_response>]]]]")
     parser.add_argument("--input-len",
                         type=int,
                         default=None,
@@ -589,11 +483,6 @@ if __name__ == "__main__":
                         action='store_true',
                         default=False,
                         help="Disable decoupled async engine frontend.")
-    parser.add_argument(
-        "--disable-detokenize",
-        action="store_true",
-        help=("Do not detokenize the response (i.e. do not include "
-              "detokenization time in the measurement)"))
     # LoRA
     parser.add_argument(
         "--lora-path",
@@ -601,45 +490,43 @@ if __name__ == "__main__":
         default=None,
         help="Path to the lora adapters to use. This can be an absolute path, "
         "a relative path, or a Hugging Face model identifier.")
-    parser.add_argument(
-        "--prefix-len",
-        type=int,
-        default=None,
-        help=f"Number of prefix tokens to be used in RandomDataset "
-        "and SonnetDataset. For RandomDataset, the total input "
-        "length is the sum of prefix-len (default: "
-        f"{RandomDataset.DEFAULT_PREFIX_LEN}) and a random context length "
-        "sampled from [input_len * (1 - range_ratio), "
-        "input_len * (1 + range_ratio)]. For SonnetDataset, "
-        f"prefix_len (default: {SonnetDataset.DEFAULT_PREFIX_LEN}) "
-        "controls how much of the input is fixed lines versus "
-        "random lines, but the total input length remains approximately "
-        "input_len tokens.")
-    # random dataset
-    parser.add_argument(
-        "--random-range-ratio",
-        type=float,
-        default=None,
-        help=f"Range ratio (default : {RandomDataset.DEFAULT_RANGE_RATIO}) "
-        "for sampling input/output length, "
-        "used only for RandomDataset. Must be in the range [0, 1) to "
-        "define a symmetric sampling range "
-        "[length * (1 - range_ratio), length * (1 + range_ratio)].",
-    )
-
-    # hf dtaset
-    parser.add_argument("--hf-subset",
-                        type=str,
-                        default=None,
-                        help="Subset of the HF dataset.")
-    parser.add_argument("--hf-split",
-                        type=str,
-                        default=None,
-                        help="Split of the HF dataset.")
 
     parser = AsyncEngineArgs.add_cli_args(parser)
     args = parser.parse_args()
     if args.tokenizer is None:
         args.tokenizer = args.model
-    validate_args(args)
+    if args.dataset is None:
+        assert args.input_len is not None
+        assert args.output_len is not None
+    else:
+        assert args.input_len is None
+    if args.enable_lora:
+        assert args.lora_path is not None
+
+    if args.backend == "vllm":
+        if args.hf_max_batch_size is not None:
+            raise ValueError("HF max batch size is only for HF backend.")
+    elif args.backend == "hf":
+        if args.hf_max_batch_size is None:
+            raise ValueError("HF max batch size is required for HF backend.")
+        if args.quantization is not None:
+            raise ValueError("Quantization is only for vLLM backend.")
+        if args.enable_lora is not None:
+            raise ValueError("LoRA benchmarking is only supported for vLLM"
+                             " backend")
+    elif args.backend == "mii":
+        if args.dtype != "auto":
+            raise ValueError("dtype must be auto for MII backend.")
+        if args.n != 1:
+            raise ValueError("n must be 1 for MII backend.")
+        if args.quantization is not None:
+            raise ValueError("Quantization is only for vLLM backend.")
+        if args.hf_max_batch_size is not None:
+            raise ValueError("HF max batch size is only for HF backend.")
+        if args.tokenizer != args.model:
+            raise ValueError("Tokenizer must be the same as the model for MII "
+                             "backend.")
+        if args.enable_lora is not None:
+            raise ValueError("LoRA benchmarking is only supported for vLLM"
+                             " backend")
     main(args)