youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[BugFix] Fix test breakages from transformers 4.45 upgrade (#8829)

Browse Source
This commit is contained in:
Nick Hill
2024-09-27 00:46:43 +01:00
committed by GitHub
parent 71d21c73ab
commit 4b377d6feb
13 changed files with 62 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/models/decoder_only/language/test_granite.py
View File

@ -3,7 +3,6 @@
Run `pytest tests/models/test_granite.py`.
"""
import pytest
import transformers
from ...utils import check_logprobs_close
@ -12,9 +11,6 @@ MODELS = [
]
# GraniteForCausalLM will be in transformers >= 4.45
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="granite model test requires transformers >= 4.45")
@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("dtype", ["bfloat16"])
@pytest.mark.parametrize("max_tokens", [64])

5
tests/models/decoder_only/vision_language/test_llava_next_video.py
View File

@ -1,7 +1,6 @@
from typing import List, Optional, Tuple, Type, overload
import pytest
import transformers
from transformers import AutoConfig, AutoModelForVision2Seq, AutoTokenizer
from vllm.multimodal.utils import (rescale_video_size, resize_video,
@ -158,8 +157,6 @@ def run_test(
)
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"size_factors",
@ -203,8 +200,6 @@ def test_models(hf_runner, vllm_runner, video_assets, model, size_factors,
)
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"sizes",

13
tests/models/decoder_only/vision_language/test_llava_onevision.py
View File

@ -1,7 +1,6 @@
from typing import List, Optional, Tuple, Type, overload
import pytest
import transformers
from transformers import (AutoConfig, AutoModelForVision2Seq, AutoTokenizer,
BatchEncoding)
@ -166,8 +165,6 @@ def run_video_test(
)
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"size_factors",
@ -211,8 +208,6 @@ def test_models(hf_runner, vllm_runner, video_assets, model, size_factors,
)
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize(
"sizes",
@ -259,7 +254,9 @@ def run_image_test(
# max_model_len should be greater than image_feature_size
with vllm_runner(model,
dtype=dtype,
max_model_len=32768,
max_num_seqs=1,
max_model_len=16384,
gpu_memory_utilization=0.98,
tensor_parallel_size=tensor_parallel_size,
distributed_executor_backend=distributed_executor_backend,
enforce_eager=True,
@ -305,8 +302,8 @@ def run_image_test(
)
@pytest.mark.skipif(transformers.__version__ < "4.45",
reason="Waiting for next transformers release")
# FIXME: Swap to a smaller model for this architecture
@pytest.mark.skip(reason="Model OOMing on CI")
@pytest.mark.parametrize("model", models)
@pytest.mark.parametrize("dtype", ["half"])
@pytest.mark.parametrize("max_tokens", [128])