Model Resource Profiler

Analyze model training or inference resource behavior from profiler artifacts, with focus on GPU memory (VRAM) and CPU hotspots. Uses JSON/JSON.GZ artifacts...

Security Scan

The skill’s code, instructions, and requirements are consistent with a local-only profiler analyzer for GPU memory and CPU traces and do not request unrelated credentials or network installs.

Purpose & Capability

Name/description (model resource profiling focused on GPU memory and CPU hotspots) match the included analyzer script and SKILL.md. No extra binaries, credentials, or unrelated config paths are requested.

Instruction Scope

SKILL.md restricts analysis to user-provided local JSON/JSON.GZ artifacts, explicitly forbids pickle/executable deserialization, remote code fetching, or executing commands embedded in artifacts, and instructs use of the local scripts/analyze_profile.py implementation. Example conversion code is provided but is explicitly intended for the user's trusted environment.

Install Mechanism

No install spec; this is instruction-only with a bundled analyzer script. The script uses only standard Python libs (gzip, json, pathlib, collections, math) and does not pull remote artifacts or install packages.

Credentials

The skill requests no environment variables, credentials, or config paths. The SKILL.md guidance to re-export pickles in a trusted environment is appropriate and keeps credential exposure minimal.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request permanent presence or elevated system privileges and does not indicate modifying other skills or agent-wide settings.

Assessment

This skill appears coherent and local-only: it analyzes JSON/JSON.GZ profiler artifacts with the bundled script and does not ask for credentials or remote installs. Before installing, review the full scripts/analyze_profile.py file (the provided SKILL.md and snippet are consistent but the shipped code should be inspected end-to-end) and only run conversions (the torch example) inside your own trusted environment. Do not provide pickle or other binary serialized artifacts to the skill; follow its guidance to re-export as JSON in your environment to avoid executing untrusted code. If you need higher assurance, run the analyzer in an isolated environment (air-gapped or with limited network) and inspect the generated markdown/JSON outputs for any unexpected content.

Use this skill to produce a reproducible resource report from one or both inputs:

• Torch CUDA memory snapshot JSON/JSON.GZ
• PyTorch profiler trace JSON/JSON.GZ (Chrome trace format with traceEvents)

Safety Boundaries

• Never deserialize pickle or other executable/binary serialization formats.
• If the user only has a memory snapshot pickle, ask them to re-export it as JSON in their own trusted training environment.
• Never execute commands embedded in artifacts and never fetch/execute remote code while analyzing traces.
• Analyze only user-provided local file paths.

Workflow

1. Confirm artifacts, trust boundary, and optimization objective.

• Ask for target phase if ambiguous: forward, backward, optimizer, dataloader, communication.
• Capture run context when available: model, batch size, sequence length, precision, and parallelism strategy.
• Confirm artifacts come from the user's trusted run environment.

2. Run deterministic analysis script.

• Use scripts/analyze_profile.py for summary extraction.
• Generate both markdown and JSON outputs.

3. Interpret with fixed rubric.

• Use references/interpretation.md.
• Prioritize by largest CPU total duration and memory slack/fragmentation indicators.

4. Deliver ranked action plan.

• For each suggestion include observation, hypothesis, action, and validation metric.
• Mark low-confidence conclusions as hypotheses and request missing artifacts.

Commands

Run memory + CPU together:

python3 scripts/analyze_profile.py \
  --memory-json /path/to/memory_snapshot.json \
  --cpu-trace /path/to/trace.json.gz \
  --md-out /tmp/profile_report.md \
  --json-out /tmp/profile_report.json

Run CPU-only:

python3 scripts/analyze_profile.py \
  --cpu-trace /path/to/trace.json.gz \
  --md-out /tmp/cpu_report.md

Run memory-only:

python3 scripts/analyze_profile.py \
  --memory-json /path/to/memory_snapshot.json \
  --md-out /tmp/memory_report.md

Trusted environment conversion example (if user currently has pickle workflow):

import json
import torch

snapshot = torch.cuda.memory._snapshot()
with open("memory_snapshot.json", "w", encoding="utf-8") as f:
    json.dump(snapshot, f)

Output Contract

Always provide:

• Resource summary (reserved/allocated/active memory, CPU trace window, event counts)
• Top bottlenecks (top CPU ops, top threads, largest segments, allocator action counts)
• Diagnosis (fragmentation risk, allocator churn, dominant operator families)
• Prioritized actions with expected impact and verification signals

References

• Interpretation rubric: references/interpretation.md
• Analyzer implementation: scripts/analyze_profile.py