You can deploy this model and start earning money today!

leibnitz-lab/
4d-llm-grid-metrics

language-model-pretraining

training-dynamics

learning-curves

Views

No views yet

4D LLM Scaling-Law Grid: Run Metrics

Per-step training metrics from a four-dimensional grid of small language-model pretraining runs, produced for the scaling-law study "Tokens-per-Parameter Coverage Is Critical for Robust LLM Scaling Law Extrapolation". The archive was moved here from the lab-v2/4d_llm_grid code repository to keep the large binary metric files out of Git.

Contents

metrics/ holds 462 gzipped CSV files (~2.4 GB), one per training run. Filenames… See the full description on the dataset page: https://huggingface.co/datasets/leibnitz-lab/4d-llm-grid-metrics.