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

[VLM] Qwen2.5-VL

Browse Source
This commit is contained in:
Roger Wang
2025-02-05 13:31:38 -08:00
committed by GitHub
parent 9a5b1554b4
commit bf3b79efb8
14 changed files with 1315 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/models/decoder_only/vision_language/test_models.py
View File

@ -121,6 +121,8 @@ VLM_TEST_SETTINGS = {
else ("half", "float")),
marks=[pytest.mark.core_model],
),
# TODO(ywang96): Move Qwen2-VL out of core models in favor of Qwen2.5-VL
# once we upgraded to transformers>=4.49.0.
"qwen2_vl": VLMTestInfo(
models=["Qwen/Qwen2-VL-2B-Instruct"],
test_type=(
@ -138,6 +140,26 @@ VLM_TEST_SETTINGS = {
image_size_factors=[(), (0.25,), (0.25, 0.25, 0.25), (0.25, 0.2, 0.15)],
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
"qwen2_5_vl": VLMTestInfo(
models=["Qwen/Qwen2.5-VL-3B-Instruct"],
test_type=(
VLMTestType.IMAGE,
VLMTestType.MULTI_IMAGE,
VLMTestType.VIDEO
),
prompt_formatter=lambda img_prompt: f"<|im_start|>User\n{img_prompt}<|im_end|>\n<|im_start|>assistant\n", # noqa: E501
img_idx_to_prompt=lambda idx: "<|vision_start|><|image_pad|><|vision_end|>", # noqa: E501
video_idx_to_prompt=lambda idx: "<|vision_start|><|video_pad|><|vision_end|>", # noqa: E501
max_model_len=4096,
max_num_seqs=2,
auto_cls=AutoModelForVision2Seq,
vllm_output_post_proc=model_utils.qwen2_vllm_to_hf_output,
image_size_factors=[(), (0.25,), (0.25, 0.25, 0.25), (0.25, 0.2, 0.15)],
marks=[pytest.mark.skipif(
TRANSFORMERS_VERSION < "4.49.0",
reason="HF model requires transformers>=4.49.0",
), pytest.mark.core_model, pytest.mark.cpu_model],
),
#### Extended model tests
"aria": VLMTestInfo(
models=["rhymes-ai/Aria"],

1
tests/models/multimodal/processing/test_common.py
View File

@ -161,6 +161,7 @@ def _test_processing_correctness(
"nvidia/NVLM-D-72B",
"Qwen/Qwen-VL-Chat",
"Qwen/Qwen2-VL-2B-Instruct",
"Qwen/Qwen2.5-VL-3B-Instruct",
"Qwen/Qwen2-Audio-7B-Instruct",
"fixie-ai/ultravox-v0_3",
])

2
tests/models/registry.py
View File

@ -264,6 +264,8 @@ _MULTIMODAL_EXAMPLE_MODELS = {
trust_remote_code=True),
"Qwen2AudioForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2-Audio-7B-Instruct"), # noqa: E501
"Qwen2VLForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2-VL-2B-Instruct"), # noqa: E501
"Qwen2_5_VLForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2.5-VL-3B-Instruct", # noqa: E501
min_transformers_version="4.49"), # noqa: E501
"UltravoxModel": _HfExamplesInfo("fixie-ai/ultravox-v0_3",
trust_remote_code=True),
# [Encoder-decoder]