Plume 256K
Plume 256K is BSC's translation research model with a 256,000-token vocabulary.
Research release
Model and training
This Plume model uses the largest vocabulary of the three research variants. BSC trains the Gemma 2B architecture from scratch on Catalan-centred parallel examples. The released card presents sentence translation in supervised and zero-shot directions. It does not establish general conversation tuning. The vocabulary change alters the embedding size as well as tokenisation.Plume 256K: official model card
Practical use
Practical guidance. Use this checkpoint to study whether broader token coverage improves the intended translation pairs. More vocabulary entries do not automatically mean better translation. Record the tokenizer and weight version together, and compare difficult names, rare words and morphologically rich sentences across the three Plume releases.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Plume256k |
| Context window | 8,192 tokens Configured maximum positions; the documented training task is sentence translation, not long-document translation. Plume256k: released model configuration |
| License | Apache License 2.0.Plume 256K: official model card |
| Model size | Gemma 2B architecture; total size varies with the vocabulary.Plume 256K: official model card |
| Architecture configuration | 18 decoder layers, 2,048 hidden width and 256,000 vocabulary entries.Plume256k: released model configuration |
Related models
Plume 128K is BSC's translation research model with a 128,000-token vocabulary.
Launch date not confirmedResearch releaseDownloadablePlume 32K is BSC's translation research model with a 32,000-token vocabulary.
Launch date not confirmedResearch releaseDownloadable