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

Revert "[V1] Scatter and gather placeholders in the model runner" (#16075)

Browse Source
This commit is contained in:
Roger Wang
2025-04-04 14:50:57 -07:00
committed by GitHub
parent f5722a5052
commit af51d80fa1
42 changed files with 942 additions and 496 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/models/decoder_only/audio_language/test_ultravox.py
View File

@ -55,10 +55,7 @@ def server(request, audio_assets):
for key, value in request.param.items()
]
with RemoteOpenAIServer(MODEL_NAME,
args,
env_dict={"VLLM_AUDIO_FETCH_TIMEOUT":
"30"}) as remote_server:
with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
yield remote_server

2
tests/models/decoder_only/vision_language/test_models.py
View File

@ -167,7 +167,7 @@ VLM_TEST_SETTINGS = {
"cherry_blossom": "<image>What is the season?", # noqa: E501
}),
multi_image_prompt="<image><image>Describe the two images in detail.", # noqa: E501
max_model_len=4096,
max_model_len=8192,
max_num_seqs=2,
auto_cls=AutoModelForImageTextToText,
vllm_runner_kwargs={"mm_processor_kwargs": {"crop_to_patches": True}}

26
tests/models/decoder_only/vision_language/test_pixtral.py
View File

@ -176,8 +176,6 @@ def test_chat(
model,
dtype=dtype,
tokenizer_mode="mistral",
load_format="mistral",
config_format="mistral",
max_model_len=max_model_len,
limit_mm_per_prompt=LIMIT_MM_PER_PROMPT,
) as vllm_model:
@ -200,14 +198,22 @@ def test_chat(
@large_gpu_test(min_gb=48)
@pytest.mark.parametrize("prompt,expected_ranges",
[(_create_engine_inputs_hf(IMG_URLS[:1]),
[PlaceholderRange(offset=11, length=494)]),
(_create_engine_inputs_hf(IMG_URLS[1:4]), [
PlaceholderRange(offset=11, length=266),
PlaceholderRange(offset=277, length=1056),
PlaceholderRange(offset=1333, length=418)
])])
@pytest.mark.parametrize(
"prompt,expected_ranges",
[(_create_engine_inputs_hf(IMG_URLS[:1]), [{
"offset": 11,
"length": 494
}]),
(_create_engine_inputs_hf(IMG_URLS[1:4]), [{
"offset": 11,
"length": 266
}, {
"offset": 277,
"length": 1056
}, {
"offset": 1333,
"length": 418
}])])
def test_multi_modal_placeholders(vllm_runner, prompt,
expected_ranges: list[PlaceholderRange],
monkeypatch) -> None:

4
tests/models/multimodal/processing/test_llava_next.py
View File

@ -92,8 +92,8 @@ def _validate_image_prompt_replacements_one(
first_placeholder = image_placeholders[0]
# NOTE: There is a BOS token
assert first_placeholder.offset == 1
assert first_placeholder.length == (
assert first_placeholder["offset"] == 1
assert first_placeholder["length"] == (
len(processed_inputs["prompt_token_ids"]) - 1) // num_imgs
except Exception as exc:

4
tests/models/multimodal/processing/test_llava_onevision.py
View File

@ -92,8 +92,8 @@ def _validate_image_prompt_replacements_one(
first_placeholder = image_placeholders[0]
assert first_placeholder.offset == 0
assert first_placeholder.length == len(
assert first_placeholder["offset"] == 0
assert first_placeholder["length"] == len(
processed_inputs["prompt_token_ids"]) // num_imgs
except Exception as exc:
failed_size_excs.append((image_size, exc))

4
tests/models/registry.py
View File

@ -277,9 +277,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
trust_remote_code=True,
hf_overrides={"architectures": ["GLM4VForCausalLM"]}), # noqa: E501
"H2OVLChatModel": _HfExamplesInfo("h2oai/h2ovl-mississippi-800m",
extras={"2b": "h2oai/h2ovl-mississippi-2b"}, # noqa: E501
max_transformers_version="4.48", # noqa: E501
transformers_version_reason="HF model is not compatible."), # noqa: E501
extras={"2b": "h2oai/h2ovl-mississippi-2b"}), # noqa: E501
"InternVLChatModel": _HfExamplesInfo("OpenGVLab/InternVL2-1B",
extras={"2B": "OpenGVLab/InternVL2-2B"}, # noqa: E501
trust_remote_code=True),