メニュー
Creates and runs parameter sweeps with omnihub-sweep using templates, sweep-dir, and optional resume. Use when the user wants to run a sweep, use omnihub-sweep, create a parameter sweep, use a sweep template, set sweep-dir, or do a dry-run sweep.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Generates SLURM job scripts with omnihub-generate-job and runs them with sbatch. Use when the user wants to generate a job, run a job, use omnihub-generate-job, sbatch, set app-config, partition, or run a single OmniHub job.
Documents OmniHub repo layout, app configs, cluster and tools config, results layout, and when to use runner manual vs torchrun. Use when the user asks about conventions, app config, cluster config, tools config, results layout, or runner manual vs torchrun.
Periodically checks the results of the current or specified OmniHub job: inspects results_dir and logs, alerts on errors, and reports success. Use when the user wants to monitor a job, check for errors in results or logs, see job status, or get a success/failure report for an OmniHub job.
Primus pretraining via OmniHub -- job generation, config overrides, MASTER_PORT, HF_HOME, GPU arch (FP8), and debugging multinode hangs. Use when the user works with applications/primus-pretrain/, pretrain_wrapper.py, Primus config, primus_turbo, or multinode Primus training.
Processes and indexes OmniHub job results with omnihub-process and omnihub-index to produce index.csv and per-job processed-data. Use when the user wants to post-process results, run omnihub-process, omnihub-index, build index.csv, or work with processed-data.
| name | sweeps |
| description | Creates and runs parameter sweeps with omnihub-sweep using templates, sweep-dir, and optional resume. Use when the user wants to run a sweep, use omnihub-sweep, create a parameter sweep, use a sweep template, set sweep-dir, or do a dry-run sweep. |
omnihub-sweep expands the template into all combinations and writes configs to sweep-dir/configurations/ (e.g. config-00000.yaml, ...).sweep-dir/jobs/ and submits it (unless --dry-run).sweep-dir/submitted.yaml. Re-run the same sweep command to skip already-submitted combinations and continue.| Argument | Description | Default |
|---|---|---|
| --cluster | hpcfund, frontier | hpcfund |
| --partitions | Partitions to sweep (space-separated) | cluster default |
| --num-nodes | List of node counts (e.g. 1 2 4) | [1] |
| --tools | Tool sets; repeat to add multiple sets (e.g. --tools omnistat --tools omnistat rocprofv3-stats) | [[]] |
| --runner | manual, torchrun (for multi-node) | — |
| --time-limit | Job time limit | 1h |
| --delay | Seconds between submissions | 60 |
| --dry-run | Generate jobs only, do not sbatch | false |
| --max-active | Max concurrent (pending+running) jobs; wait until below before submitting more | 10 |
| --platform | apptainer, docker | apptainer |
| --rocm-version | ROCm version | 7.1.0 |
| --include-nodelist, --exclude-nodelist | Node filters | — |
YAML with list values; each combination becomes one config:
entrypoint: /app/vllm/benchmarks/benchmark_latency.py
model: meta-llama/Llama-3.1-8B-Instruct
tensor_parallel_size: 1
input_len: [32, 64]
output_len: [128, 256]
batch_size: 8
This yields 4 configs (2×2). Sweep then multiplies by partitions × num_nodes × tool sets.
sweep-dir/
configurations/ # config-00000.yaml, config-00001.yaml, ...
jobs/ # job-<partition>-<nodes>-<toolset>-<config>.sh
submitted.yaml # list of { partition, num_nodes, tools, config, job_id } for resume
mkdir -p sweep-vllm
./omnihub-sweep --omnihub-dir $PWD --sweep-dir ./sweep-vllm \
--template vllm-latency-template.yaml \
--partitions mi2104x mi2508x \
--tools omnistat --tools omnistat rocprofv3-stats
Dry-run (generate only):
./omnihub-sweep --omnihub-dir $PWD --sweep-dir ./sweep-vllm \
--template vllm-latency-template.yaml --dry-run
Re-run the same command (same sweep-dir and parameters). Already-submitted (partition, num_nodes, tools, config) entries are read from submitted.yaml and skipped.