First 70B model released with all training epochs and data
11 hours ago
- #Korean
- #LLM
- #checkpoints
- First release of intermediate checkpoints from the Tri family (0.5B, 1.9B, 7B) among Korean-targeted LLMs.
- Checkpoints released at regular token intervals: ~20B (0.5B), ~40B (1.9B), ~160B (7B & 70B).
- Includes 0.5B and 1.9B runs originally for system bring-up, now useful for small-scale training analysis.
- Links to checkpoints: Tri-0.5B, Tri-1.9B, Tri-7B, Tri-70B (SFT Preview).
- Usage example provided with `AutoModelForCausalLM` and `AutoTokenizer` from Hugging Face.