Salamandra 7B Instruct
The original Salamandra 7B Instruct is BSC's first conversational adaptation of its 7B multilingual base.
Research release
Model and training
The release uses 273,000 instruction examples, with a focus on Catalan, English and Spanish. BSC labels it a proof of concept and states that it has not undergone RLHF alignment. It retains the base model's 32-layer decoder and 8,192-token window. This record covers the original release, not the later 2606 update.Salamandra 7B Instruct: official model card
Practical use
Practical guidance. Keep this checkpoint when reproducing early Salamandra results or comparing changes in post-training. New deployments should compare the 2606 model on the same prompts. Use a fixed chat template and decoding settings so that prompt formatting does not obscure the effect of the weight changes.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | salamandra-7b-instructsalamandra-7b-instruct-fp8salamandra-7b-instruct-gptq |
| Context window | 8,192 tokens Maximum sequence length documented for these weights; the prompt and generated text share the window. Salamandra 7B Instruct: official model card |
| License | Apache License 2.0.Salamandra 7B Instruct: official model card |
| Model size | 7,768,117,248 parameters.Salamandra 7B Instruct: official model card |
Related models
- Salamandra 7B
Barcelona Supercomputing Center · Salamandra
Seven-billion-parameter multilingual base model that forms the language backbone of later Salamandra adaptations.
Launch date not confirmedDownloadable Salamandra 7B Instruct 2606 updates BSC's conversational model with longer context and preliminary preference alignment.
Launch date not confirmedDownloadable