rinnaBaku
Gemma 2 Baku 2B Instruct
Baku 2B Instruct adds Google's Gemma instruction chat vector to rinna's Japanese-adapted base, then applies odds-ratio preference optimization.
Merging excludes the embedding layer
The documented formula uses a chat-vector multiplier of 1.0 and omits the embedding layer. The release retains the Gemma 2 instruction tokenizer and conversation format.Gemma 2 Baku 2B Instruct official model card
Distinguish a small assistant from the base checkpoint
Practical guidance. Use the instruction release when evaluating task-following behavior without additional tuning. Compare short outputs and unsupported requests; compact size alone says little about the reliability of an answer.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | gemma-2-baku-2b-it |
| License | Gemma terms.Gemma 2 Baku 2B Instruct official model card |
| Ancestry | Baku 2B plus the difference between Google's Gemma 2 2B Instruct and base weights.Gemma 2 Baku 2B Instruct official model card |
Related models
- Gemma 2 Baku 2B
rinna ยท Baku
Baku 2B continues pretraining Google's Gemma 2 2B on about 80B Japanese and English tokens.
Launch date not confirmedResearch releaseDownloadable