메뉴
Prepare and validate Parameter Golf record folders: self-contained train_gpt.py, README.md, submission.json, FineWeb SP1024 BPB accounting, artifact-size logging, run logs, and PR-ready folder hygiene.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when provisioning Prime Intellect GPU compute, managing pods/disks/sandboxes, running hosted RL training via prime lab, installing or publishing RL environments, or exposing local services via Prime Tunnel. Covers the `prime` CLI (PyPI: prime) for all Prime Intellect platform operations.
Shared Python package for Science Skills, currently containing http_client -- a unified HTTP client with rate limiting, retries, and exponential backoff. Not a standalone agent skill. Do not invoke directly.
Retrieve and analyze AlphaFold predicted structures for a protein. Use when the user provides a specific UniProt Accession ID and wants structural confidence metrics (pLDDT), domain boundary analysis, or disorder assessment. Do not use if the user only has a protein name, gene name, or amino acid sequence — ask for a UniProt ID first.
Query the ChEMBL database for bioactive molecules, drug targets, bioactivity data, approved drugs, and chemical structures. Use when the user asks about compounds, targets, IC50/Ki values, drug mechanisms, or structure searches.
Query ClinicalTrials.gov via APIv2. Use when you want to search for trials by condition, drug, location, status, or phase; retrieve trial details by NCT ID; check eligibility/inclusion criteria; count trials across conditions or time periods; identify a sponsor's trial portfolio; find recruiting trials for patient matching.
Use when needing clinical significance, pathogenicity classifications (e.g., Pathogenic, Benign, VUS), clinical evidence rationales, or finding "hard positive" benchmark controls for human genomic variants.
| name | parameter-golf-submission |
| description | Prepare and validate Parameter Golf record folders: self-contained train_gpt.py, README.md, submission.json, FineWeb SP1024 BPB accounting, artifact-size logging, run logs, and PR-ready folder hygiene. |
| tags | ["parameter-golf","competition","fineweb","bpb","model-craft","submission"] |
Use this skill when creating or reviewing a Parameter Golf submission folder, independent of the cloud provider used for the run.
A submission folder must contain:
records/<track>/<submission-name>/
README.md
submission.json
train_gpt.py
train.log # after a real run
train_gpt.py must compile and run from inside this folder in a clean Parameter Golf checkout.
16,000,000 decimal bytes.val_bpb).Before running:
python -m py_compile train_gpt.py passes.torch, numpy, sentencepiece, etc.).DATA_PATH and TOKENIZER_PATH are env-configurable.fineweb_train_*.bin and fineweb_val_*.bin with the Parameter Golf binary header format.final_model.int8.ptz.final_int8_zlib_roundtrip_exact.The README must include:
Use actual values after the run, not placeholders:
{
"run_name": "...",
"author": "...",
"github_id": "...",
"track": "track_10min_16mb or track_non_record_16mb",
"val_bpb": 1.2345,
"val_loss": 2.1234,
"artifact_size_bytes": 12345678,
"command": "...",
"status": "completed"
}
Add architecture fields as useful, but avoid claiming record eligibility unless the log proves it.
After a run, extract these lines:
grep -E "final_int8_zlib_roundtrip_exact|Total submission size int8\+zlib|stopping_early|train_time|model_params" train.log
Update:
submission.json.val_bpbsubmission.json.val_losssubmission.json.artifact_size_bytesUse precise status:
prepared_pending_run: folder created, no real run yetsmoke_passed: short/non-final run passedcompleted_non_record: full run but not leaderboard-valid or not SOTAcompleted_record_candidate: 8xH100 10-minute compliant run with full log and artifact under capfailed: include failure reason and last good checkpoint/log linefrom src...) not present in record folder.train.log from logs/<RUN_ID>.txt.