What it is
Three intermediate checkpoints of a 7B custom LLM released openly for reproducibility and training-dynamics analysis. 3 raw intermediate checkpoints from a 7B custom LLM at 110k, 115k, and 162k steps, covering ~98B, ~102B, and ~144B tokens; each includes weights, config, and tokenizer under Apache-2.0.
Key results
- 3 checkpoints at 110k, 115k, and 162k steps
- ~98B, ~102B, and ~144B tokens seen
- Each folder has weights, config, and tokenizer
Useful for studying LLM training dynamics, reproducibility, and checkpoint-to-checkpoint behavior.