biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V1] Add V1 support of Qwen2-VL (#12128)

Browse Source 
Signed-off-by: Roger Wang <ywang@roblox.com>
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: imkero <kerorek@outlook.com>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Roger Wang
2025-01-19 03:52:13 -08:00
committed by GitHub
parent edaae198e7
commit 81763c58a0
9 changed files with 291 additions and 84 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
tests/models/decoder_only/vision_language/test_qwen2_vl.py
View File

@@ -105,7 +105,7 @@ def batch_make_image_embeddings(
pixel_values = preprocess_result["pixel_values"]
image_grid_thw = preprocess_result["image_grid_thw"]
# pixel values to embeddinds & grid_thws
# pixel values to embeddings & grid_thws
with torch.no_grad():
visual = llm.llm_engine.model_executor.driver_worker. \
model_runner.model.visual
@@ -124,11 +124,10 @@ def batch_make_image_embeddings(
for image_batch in image_batches_:
cur_batch_image_count = len(image_batch)
merge_size = image_processor.merge_size
cur_batch_embed_len = sum([
grid_thw.prod() // merge_size // merge_size
cur_batch_embed_len = sum(
grid_thw.prod(-1) // merge_size // merge_size
for grid_thw in image_grid_thw[image_counter:image_counter +
cur_batch_image_count]
])
cur_batch_image_count])
result.append({
"image_embeds":
@@ -187,7 +186,7 @@ def batch_make_video_embeddings(
pixel_values = preprocess_result["pixel_values_videos"]
video_grid_thw = preprocess_result["video_grid_thw"]
# pixel values to embeddinds & grid_thws
# pixel values to embeddings & grid_thws
with torch.no_grad():
visual = llm.llm_engine.model_executor.driver_worker.\
model_runner.model.visual
@@ -206,11 +205,10 @@ def batch_make_video_embeddings(
for video_batch in video_batches_:
cur_batch_video_count = len(video_batch)
merge_size = image_processor.merge_size
cur_batch_embed_len = sum([
grid_thw.prod() // merge_size // merge_size
cur_batch_embed_len = sum(
grid_thw.prod(-1) // merge_size // merge_size
for grid_thw in video_grid_thw[video_counter:video_counter +
cur_batch_video_count]
])
cur_batch_video_count])
result.append({
"video_embeds":