Llama 3.1 8B TFree-HAT DPO
The 8B TFree-HAT DPO checkpoint adds direct preference optimization to the English-and-German instruction model.
Research release
Model design
Its ancestry is the Llama 3.1 8B backbone adapted to Hierarchical Autoregressive Transformer tokenization. Preference training focuses on helpfulness and following instructions. Realized inference speed depends on the serving implementation.Llama 3.1 8B TFree-HAT DPO model card
Practical use
Practical guidance. Compare DPO with the SFT checkpoint on ambiguous instructions and refusal behavior. A preference-trained response can be more persuasive without being more accurate. Keep research weight terms and the inherited Llama ancestry visible when evaluating a deployment path.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Aleph-Alpha/llama-3_1-8b-tfree-hat-dpo |
| License | Open Aleph License for non-commercial research and education.Llama 3.1 8B TFree-HAT DPO model card |
Related models
- Llama 3.1 8B TFree-HAT BASE
Aleph Alpha · TFree-HAT
Tokenizer-free Llama 3.1 8B research adaptation with continued pretraining in English and German.
Launch date not confirmedResearch releaseDownloadable - Llama 3.1 8B TFree-HAT SFT
Aleph Alpha · TFree-HAT
Tokenizer-free Llama 3.1 8B adaptation with supervised instruction training in English and German.
Launch date not confirmedResearch releaseDownloadable