mem_profile

mem_profile is a Runtime diagnostics & profiling skill for AI agents, published by NVIDIA-BioNeMo in the boltz-cp repository on GitHub. Skillspace rates mem_profile F (45/100) and the safety scan rated it risky; the rating is based on static analysis only.

Risky. This skill has patterns associated with unsafe behaviour. Read the source carefully before running it.

Memory-profile a context-parallel (CP) inference (or training) workflow with the PyTorch CUDA caching-allocator history, then attribute the top-N memory peaks to specific modules and lines of code. Wraps the end-to-end forward in torch.cuda.memory._record_memory_history() + _dump_snapshot() (the same mechanism as Boltz2's CUDAMemoryProfile Lightning callback) under a torchrun launcher that writes one snapshot per rank, then runs a stdlib analyzer (mem_profile_analysis.py) that replays the allocation timeline, finds the distinct peaks, and emits a markdown report with clickable file:line links to the call sites holding memory at each peak — sorted by peak, then by contribution. Use once a CP workflow runs end-to-end and you need to find the largest token count that fits and which module is the memory bottleneck.

Grade

F

Quality

45 / 100

Safety

Risky

GitHub stars

51

Unproven: rated by static analysis only, not yet compared head-to-head.

Our two independent classifiers differ on this skill's category, so treat the placement as provisional.

This rating describes the version of the repository we crawled; it may have changed since.

Skillspace indexes and rates this skill; it does not host or endorse it. Run third-party skills at your own risk.