Ǎguila 7B
Ǎguila 7B is BSC's earlier Catalan, Spanish and English language adaptation of TII's Falcon 7B.
Research release
Model and training
BSC replaced Falcon's tokenizer, adjusted its embedding layer and continued training on 26 billion tokens. The adapted vocabulary has 50,257 tokens. The card identifies the Language Technologies Unit as developer and describes a base causal language model intended for further task tuning. It is an AINA-era derivative, not a new model trained entirely from scratch. BSC marks this model as deprecated and no longer maintained or supported. Its weights remain downloadable.Àguila 7B: official model card
Practical use
Practical guidance. Use Ǎguila to study language adaptation before the Salamandra programme. Compare it with its Falcon parent and with a from-scratch multilingual model on the same tasks. Preserve its own tokenizer during evaluation; mixing a parent tokenizer with the adapted embeddings changes the input representation.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | aguila-7b |
| License | Apache License 2.0.Àguila 7B: official model card |
| Model size | 7B model class.Àguila 7B: official model card |
Related models
FLOR 6.3B is BSC's larger BLOOM-derived language model for Catalan, Spanish and English.
Launch date not confirmedRestricted accessDownloadable- 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