Llama TFree-HAT Pretrained 7B DPO
This 7B tokenizer-free model was pretrained from scratch, then instruction- and preference-trained in English and German. Its Llama prefix reflects use of Llama 3.3 in data filtering, rather than initialization from Llama weights.
Research release
Model design
The checkpoint is adapted from the TFree-HAT pretrained base and optimized for helpfulness and instruction following. The card says it is not optimized for code generation or mathematics. It requires the model-specific inference implementation.Llama TFree-HAT Pretrained 7B DPO model card
Practical use
Practical guidance. Use this checkpoint when studying a tokenizer-free model without inherited Llama backbone weights. Compare it with the Llama-initialized releases using the same language tasks. Document data-filtering ancestry separately from parameter initialization so the model's origin is clear.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Aleph-Alpha/llama-tfree-hat-pretrained-7b-dpo |
| License | Open Aleph License for non-commercial research and education.Llama TFree-HAT Pretrained 7B DPO model card |
| Architecture | Character-level encoding and decoding around a word-level backbone.Llama TFree-HAT Pretrained 7B DPO model card |
| Public use | Non-commercial research and education.Llama TFree-HAT Pretrained 7B DPO model card |
Related models
- Llama 3.1 8B TFree-HAT DPO
Aleph Alpha ยท TFree-HAT
English-and-German TFree-HAT instruction model further trained with direct preference optimization.
Launch date not confirmedResearch releaseDownloadable