// 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.
| 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.Run Parameter Golf competition submissions on RunPod GPU Pods. Covers required operator inputs, RunPod pod specs, FineWeb SP1024 data caching, record-folder hygiene, torchrun launch commands, monitoring, artifact-size checks, and result collection.
Hands-on implementation template and API reference for writing, tuning, debugging, and benchmarking Triton GPU kernels. Covers the full triton.language API surface, autotuning patterns, profiling workflows, and production integration.
Hands-on implementation template and API reference for writing, tuning, debugging, and benchmarking Triton GPU kernels. Covers the full triton.language API surface, autotuning patterns, profiling workflows, and production integration.
Tencent AngelSlim — accessible, comprehensive, and efficient toolkit for large model compression. Quantization (FP8/INT4/NVFP4/1.25-bit), pruning, speculative decoding (Eagle3), and diffusion model compression.
DistilQwen2.5 — Alibaba's industrial practices for training distilled open lightweight language models. Knowledge distillation from Qwen2.5 72B into smaller 0.5B-7B models.
Intel Neural Compressor — SOTA low-bit LLM quantization (INT8/FP8/INT4/NVFP4), sparsity, pruning, and distillation for PyTorch, TensorFlow, and ONNX Runtime.