rinnaYouko
Llama 3 Youko 70B
Youko 70B continues pretraining Meta Llama 3 70B on 5B Japanese and English tokens.
Research release
A bilingual training stage follows Meta's base
The model retains 80 layers with hidden width 8,192. rinna's contribution is continued language training. The name Youko refers to a fox-like creature in Japanese folklore.Llama 3 Youko 70B official model card
Keep adaptation data separate from original pretraining
Practical guidance. The stated five-billion-token stage measures rinna's additional training, not the total data behind Llama 3. Use that boundary when comparing training histories across model families.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | llama-3-youko-70bllama-3-youko-70b-gptq |
| Ancestry | meta-llama/Meta-Llama-3-70B.Llama 3 Youko 70B official model card |
| License | Meta Llama 3 Community License.Llama 3 Youko 70B official model card |
Related models
- Llama 3 Youko 70B Instruct
rinna · Youko
Youko 70B Instruct combines supervised fine-tuning with a chat vector from Meta's Llama 3 instruction model.
Launch date not confirmedDownloadable - Llama 3 Youko 8B
rinna · Youko
Youko 8B continues pretraining Meta Llama 3 8B on 22B Japanese and English tokens.
Launch date not confirmedResearch releaseDownloadable