A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 6, 23:00 UTC
Fix Qwen2.5-VL test for Transformers v5 (#36532)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Mar 10, 2026 at 12:05 UTC
c88510083b8d6b4fa7a42ae29bc27ff6adc181ee
2 files changed
+2
-6
examples/pooling/classify/vision_classification_online.py
+1
-1
@@ -8,7 +8,7 @@ NOTE:
8
--runner pooling \
9
--max-model-len 5000 \
10
--limit-mm-per-prompt.video 1 \
11
- --hf-overrides '{"text_config": {"architectures": ["Qwen2_5_VLForSequenceClassification"]}}'
11
+ --hf-overrides '{"architectures": ["Qwen2_5_VLForSequenceClassification"]}'
12
"""
13
14
import argparse
tests/entrypoints/pooling/classify/test_online_vision.py
+1
-5
@@ -12,11 +12,7 @@ from vllm.multimodal.utils import encode_image_url, fetch_image
12
MODEL_NAME = "muziyongshixin/Qwen2.5-VL-7B-for-VideoCls"
13
MAXIMUM_VIDEOS = 1
14
15
-HF_OVERRIDES = {
16
- "text_config": {
17
- "architectures": ["Qwen2_5_VLForSequenceClassification"],
18
- },
19
-}
15
+HF_OVERRIDES = {"architectures": ["Qwen2_5_VLForSequenceClassification"]}
16
input_text = "This product was excellent and exceeded my expectations"
17
image_url = "https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/cat_snow.jpg"
18
image_base64 = {"url": encode_image_url(fetch_image(image_url))}