Empirical GPU training timing measurements collected to calibrate analytical (roofline-based) LLM training time estimators. The dataset captures real step times across multiple GPU architectures, model families, and configurations, along with the ratio of measured to predicted times (the correction factor).
Roofline models tend to significantly underestimate actual training time, particularly for small models on large GPUs where memory bandwidth… See the full description on the dataset page:
https://huggingface.co/datasets/osteele/llm-calibration-db.
