[Bugfix][Qwen2.5-VL] Honor video fps for temporal M-RoPE (#47736)

Signed-off-by: Ting Sun <suntcrick@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Ting SUN
2026-09-22 05:56:12 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 5abf85d27a
commit a33b3bac5d
7 changed files with 117 additions and 8 deletions
@@ -151,10 +151,12 @@ VLM_TEST_SETTINGS = {
img_idx_to_prompt=lambda idx: "<|vision_start|><|image_pad|><|vision_end|>",
video_idx_to_prompt=lambda idx: "<|vision_start|><|video_pad|><|vision_end|>",
enforce_eager=False,
needs_video_metadata=True,
max_model_len=4096,
max_num_seqs=2,
auto_cls=AutoModelForImageTextToText,
vllm_output_post_proc=model_utils.qwen2_vllm_to_hf_output,
patch_hf_runner=model_utils.qwen3_vl_patch_hf_runner,
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],
),
@@ -4,10 +4,10 @@
import pytest
from vllm.assets.image import ImageAsset
from vllm.multimodal.video import sample_frames_from_video
from vllm.platforms import current_platform
from ....conftest import VIDEO_ASSETS
from .vlm_utils.builders import sample_frames_with_video_metadata
models = ["Qwen/Qwen2.5-VL-3B-Instruct"]
target_dtype = "bfloat16"
@@ -81,7 +81,9 @@ def test_qwen2_5_vl_evs_functionality(
# Sample frames from video assets
sampled_vids = [
sample_frames_from_video(asset.np_ndarrays, num_frames)
sample_frames_with_video_metadata(
(asset.np_ndarrays, asset.metadata), num_frames
)
for asset in video_assets
]
@@ -145,7 +147,9 @@ def test_qwen2_5_vl_evs_batched_videos(
monkeypatch.setenv("VLLM_USE_BYTECODE_HOOK", "1" if use_bytecode_hook else "0")
# Sample frames from video assets
sampled_vids = [
sample_frames_from_video(asset.np_ndarrays, num_frames)
sample_frames_with_video_metadata(
(asset.np_ndarrays, asset.metadata), num_frames
)
for asset in video_assets
]
@@ -144,7 +144,7 @@ MODEL_CONFIGS: dict[str, VitCudagraphTestConfig] = {
"<|vision_start|><|video_pad|><|vision_end|>"
"Describe this video in one sentence."
),
needs_video_metadata=False,
needs_video_metadata=True,
marks=[pytest.mark.core_model],
),
"kimi_vl": VitCudagraphTestConfig(