Skip to content

inference: add CLI overrides for fps/total_pixels and vLLM memory knobs to prevent CUDA OOM in video runs #83

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

Open
paularamo wants to merge 4 commits into
base: main
Choose a base branch
from
Open

inference: add CLI overrides for fps/total_pixels and vLLM memory knobs to prevent CUDA OOM in video runs #83

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
19481c9
inference: add CLI overrides for fps/total_pixels and vLLM memory knobs
paularamo Nov 14, 2025
0919ddf
avoid CUDA OOM
paularamo Nov 18, 2025
7f6baa2
chore: fix EOF newline and apply CLI override patch
paularamo Nov 25, 2025
925275b
chore: mark inference script as executable
paularamo Nov 25, 2025
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
22 changes: 21 additions & 1 deletion scripts/inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -148,6 +148,20 @@ def main():
type=str,
help="Output directory for debugging",
)
# Adding the next two arguments to improve performance on the GPU
parser.add_argument(
"--gpu-memory-utilization",
type=float,
default=None,
help="Target fraction of GPU memory vLLM can use for model + KV cache",
)
parser.add_argument(
"--max-model-len",
type=int,
default=None,
help="Maximum sequence length for sizing KV cache",
)
Comment thread
paularamo marked this conversation as resolved.

args = parser.parse_args()

images: list[str] = args.images or []
Expand Down Expand Up @@ -204,6 +218,8 @@ def main():
revision=args.revision,
limit_mm_per_prompt={"image": len(images), "video": len(videos)},
enforce_eager=True,
gpu_memory_utilization=args.gpu_memory_utilization,
max_model_len=args.max_model_len,
)

# Process inputs
Expand Down Expand Up @@ -239,14 +255,18 @@ def main():
"mm_processor_kwargs": video_kwargs,
}
outputs = llm.generate([llm_inputs], sampling_params=sampling_params)

print(SEPARATOR)
full_texts = []
for output in outputs[0].outputs:
output_text = output.text
full_texts.append(output_text)
print("Assistant:")
print(textwrap.indent(output_text.rstrip(), " "))
print(SEPARATOR)

result, _ = extract_tagged_text(output_text)
result, _ = extract_tagged_text(full_texts[-1])

if args.verbose and result:
pprint_dict(result, "Result")

Expand Down