Salamandra 7B FC 2610 Evo
Salamandra 7B FC 2610 Evo is BSC's experimental tool-use model trained with reinforcement learning.
Restricted access
Model and training
The card names Salamandra 7B Instruct 2606 as the starting model. Training uses GRPO with EnvScaller and internal Wikipedia and open-data environments. BSC states that comprehensive capability and safety evaluation is not complete. The repository is manually gated, so this is a restricted research release rather than an unrestricted download.Salamandra 7B FC 2610 Evo: official model card
Practical use
Practical guidance. Use this checkpoint for research into tool interaction and reinforcement learning. Keep environment-specific success separate from performance on unfamiliar tools. The public documentation does not justify assuming broad autonomous-agent competence. Reproduce the input format and inspect actions, failed calls and final answers together.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable |
| Model identifiers | salamandra-7b-fc-2610-evo |
| License | Apache License 2.0.Salamandra 7B FC 2610 Evo: official model card |
| Model size | 7B model class.Salamandra 7B FC 2610 Evo: official model card |
Related models
- Salamandra 7B FC 2607
Barcelona Supercomputing Center · Salamandra
Salamandra 7B FC 2607 adapts BSC's 2606 instruction model for function calling in a smaller model class.
Launch date not confirmedRestricted accessDownloadable Salamandra 7B Instruct 2606 updates BSC's conversational model with longer context and preliminary preference alignment.
Launch date not confirmedDownloadable