CUDA out of memory during vLLM CUDA graph capture
CUDA out of memory
The first OOM appears while capturing or warming up CUDA graphs after weights load. Captured graphs can require additional GPU memory.
Source checked against vLLM v0.32.0. Log wording and available flags may vary by release.
First action
Repeat the same launch with --enforce-eager as a control. If the OOM disappears, use the capture settings supported by your release to test smaller capture sizes before making a performance tradeoff permanent.
Verify the change. Compare startup, peak memory, and serving latency with the same workload. Keep the earlier traceback if the eager run still fails.
Diagnose your vLLM log offline
Synthetic demonstration samples
Synthetic demonstration samples. Not customer telemetry.
Upload your log file. Large files are processed in your browser. Only selected evidence is submitted when you run diagnosis.
Your logs are processed entirely in your browser and are never uploaded by this tool.
Add host kernel / dmesg or nvidia-smi context (optional)
100% local browser engine (same deterministic core as the Open VSX & VS Code extension). 0 bytes of log data uploaded.
Handling a production incident? Replay it with your team.