Skip to content

[VLM] Create empty videos vector with batch #3129

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Wovchena merged 6 commits into from
Jan 8, 2026
Merged

[VLM] Create empty videos vector with batch #3129

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
83771c5
Create empty videos vector with batch
as-suvorov Dec 18, 2025
6c61044
Add generic ov image placeholder
as-suvorov Dec 18, 2025
a7c705b
Merge branch 'master' into as/vlm_via_cb_no_video_with_batch
Wovchena Dec 29, 2025
1262db0
Merge branch 'master' into as/vlm_via_cb_no_video_with_batch
yatarkan Jan 5, 2026
7a91e87
Merge branch 'master' into as/vlm_via_cb_no_video_with_batch
as-suvorov Jan 7, 2026
13f98b2
Merge branch 'master' into as/vlm_via_cb_no_video_with_batch
Wovchena Jan 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 7 additions & 3 deletions samples/python/visual_language_chat/milebench_eval_vlm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -153,12 +153,16 @@ def __getitem__(self, idx):
context += choice_str

img_num = len(ann["task_instance"]["images_path"])
qwen2_vl_image_placeholder = "<|vision_start|><|image_pad|><|vision_end|>"

def idx_to_ov_image_placeholder(idx: int) -> str:
return f"<ov_genai_image_{idx}>"
Comment on lines +157 to +158
Copy link

Copilot AI Dec 29, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The function name idx_to_ov_image_placeholder is verbose. Consider renaming it to format_image_placeholder or get_image_placeholder for better readability.

Copilot uses AI. Check for mistakes.

for i in range(img_num):
rmv_txt = "{image#%d}" % (i + 1)
rmv_tbl = "{table#%d}" % (i + 1)
context = context.replace(rmv_txt, qwen2_vl_image_placeholder)
context = context.replace(rmv_tbl, qwen2_vl_image_placeholder)
image_placeholder = idx_to_ov_image_placeholder(i)
context = context.replace(rmv_txt, image_placeholder)
context = context.replace(rmv_tbl, image_placeholder)

task_instruction_id = ann["task_instruction_id"]
context_str = task_instructions[task_instruction_id] + "\n" + context
Expand Down
4 changes: 3 additions & 1 deletion src/cpp/src/continuous_batching/pipeline_base.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -232,7 +232,9 @@ ContinuousBatchingPipeline::IContinuousBatchingPipeline::generate(
const std::vector<std::vector<ov::Tensor>>& images_vector,
const std::vector<GenerationConfig>& sampling_params,
const StreamerVariant& streamer) {
return generate(prompts, images_vector, {{}}, sampling_params, streamer);
// empty videos batch size should match prompt batch size
const std::vector<std::vector<ov::Tensor>> empty_videos_vector(prompts.size());
return generate(prompts, images_vector, empty_videos_vector, sampling_params, streamer);
}

std::vector<VLMDecodedResults>
Expand Down
Loading