6 experiments provide IsoFLOP curve data standardized to a common schema. This curated collection supports analysis of systematic biases in scaling law fits, as detailed in the associated arXiv paper 2603.22339. The data originates from sources including ml_scalefit, epochai_chinchilla, and llama_3.
Use Cases
- Compare IsoFLOP curve shapes across different data sources like ml_scalefit and epochai_chinchilla to identify systematic biases.
- Analyze the source field to validate scaling law fits from experiments like llama_3 and marin_202603.
- Standardize performance metrics from the six experiments using the common schema for meta-analysis of training efficiency.
Strengths
- Data from 6 distinct experiments enables cross-source validation.
- Standardized schema facilitates direct comparison of curves from different sources.
Limitations
- Unknown row count and column details limit assessment of analytical scope.
- Sample data and file formats are unavailable, hindering preliminary inspection.
Provenance
- Source
- open-athena on Hugging Face.
- Collection Method
- Curated collection extracted and transformed via Open-Athena/scaling-law-analysis code.
- Time Range
- null
- Freshness
- Last updated 2026-03 25.
- Geography
- null