[Model][VLM] Add Qwen2.5-Omni model support (thinker only) (#15130)

Signed-off-by: fyabc <suyang.fy@alibaba-inc.com>
Signed-off-by: Roger Wang <ywang@roblox.com>
Co-authored-by: Roger Wang <136131678+ywang96@users.noreply.github.com>
Co-authored-by: Roger Wang <ywang@roblox.com>
Co-authored-by: Xiong Wang <wangxiongts@163.com>
This commit is contained in:
Yang Fan
2025-04-18 23:14:36 -07:00
committed by GitHub
co-authored by Roger Wang Roger Wang Xiong Wang
parent 5c9121203c
commit 2c1bd848a6
23 changed files with 1855 additions and 85 deletions
@@ -139,6 +139,23 @@ 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_omni": VLMTestInfo(
models=["Qwen/Qwen2.5-Omni-7B"],
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_bos|><|IMAGE|><|vision_eos|>", # noqa: E501
video_idx_to_prompt=lambda idx: "<|vision_bos|><|VIDEO|><|vision_eos|>", # 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.core_model, pytest.mark.cpu_model],
),
#### Extended model tests
"aria": VLMTestInfo(
models=["rhymes-ai/Aria"],
@@ -280,6 +280,7 @@ def _test_processing_correctness_mistral(
"Qwen/Qwen2-VL-2B-Instruct",
"Qwen/Qwen2.5-VL-3B-Instruct",
"Qwen/Qwen2-Audio-7B-Instruct",
"Qwen/Qwen2.5-Omni-7B",
"Skywork/Skywork-R1V-38B",
"fixie-ai/ultravox-v0_5-llama-3_2-1b",
"openai/whisper-large-v3",
+2
View File
@@ -362,6 +362,8 @@ _MULTIMODAL_EXAMPLE_MODELS = {
"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
"Qwen2_5OmniModel": _HfExamplesInfo("Qwen/Qwen2.5-Omni-7B", # noqa: E501
min_transformers_version="4.52"), # noqa: E501
"SkyworkR1VChatModel": _HfExamplesInfo("Skywork/Skywork-R1V-38B"),
"SmolVLMForConditionalGeneration": _HfExamplesInfo("HuggingFaceTB/SmolVLM2-2.2B-Instruct"), # noqa: E501
"UltravoxModel": _HfExamplesInfo("fixie-ai/ultravox-v0_5-llama-3_2-1b", # noqa: E501