القائمة
Compare experiment results across servers — differing hyperparameters and final metrics from local wandb data, with dashboard links. Use for "which ablation won" or comparing runs of a project.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Find what's eating disk on full lab servers — old checkpoints, wandb and cache junk — and produce a deletion plan. Report-only; deletion needs per-path human approval. Use when a disk is full or the user asks to free space.
Morning briefing across all lab servers — what finished or crashed overnight, which GPUs are free, which disks are filling up. Use when the user asks for a briefing, "what happened overnight", or overall lab/server status.
Launch a training run on the best lab server — picks a host by data locality and free GPUs, verifies code and environment, starts it in a detached tmux session. Use when the user wants to run, launch, or train something on the servers.
Reproduce a working project environment from one lab server on another — code at the same commit, a CUDA-compatible python env, and a smoke test. Use for "make this run on host B too" or setting up a new server.
Check on the user's running experiments across all servers — progress, loss, ETA, stalled or crashed. Use for "how are my runs", "is training still going", or checking a specific experiment.
Launch a hyperparameter sweep distributed across free GPUs on multiple lab servers, one detached tmux session per config. Use when the user wants to run several variants of an experiment at once.
| name | compare-runs |
| description | Compare experiment results across servers — differing hyperparameters and final metrics from local wandb data, with dashboard links. Use for "which ablation won" or comparing runs of a project. |
Input: a project name/dir (or infer from context). All read-only.
Find the runs: list_sessions cwds point at project dirs;
list_wandb_runs on each host+dir that has the project lists recent runs
with dashboard URLs.
Pull each run's local records (newest ~10 unless told otherwise) via
run_command:
cat '<run>/files/wandb-summary.json'sed -n '1,80p' '<run>/files/config.yaml'Note: wandb-summary.json exists even for offline/unsynced runs — that's
why we read files instead of the wandb API; no key needed.
Build the comparison: run (wandb id or dir date) · host · only the hyperparams that DIFFER between runs (drop identical ones) · the key final metrics the summaries share (loss/acc/mAP/…) · state · wandb URL when found.
The table, then: the best run by the metric that matters (ask if ambiguous), and flag runs that crashed too early to be comparable. If the same config ran on different hosts with different results, call that out — it usually means code or data skew (suggest /sync-code).