Plume 128K
Plume 128K is BSC's translation research model with a 128,000-token vocabulary.
Research release
Model and training
The middle Plume release uses a 128,000-token vocabulary with the Gemma 2B architecture. BSC trains it from scratch using only Catalan-centred parallel data. The study covers Catalan with Spanish, French, Italian, Portuguese, Galician, German, English and Basque. This is sentence-level translation research, not an instruction-tuned Gemma assistant.Plume 128K: official model card
Practical use
Practical guidance. This variant is useful for examining the trade-off between vocabulary size and tokenised sentence length. Compare translation adequacy and resource use with both smaller and larger vocabularies. Do not interpret 128k as the model's context window or as evidence of document-level translation training.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Plume128k |
| Context window | 8,192 tokens Configured maximum positions; the documented training task is sentence translation, not long-document translation. Plume128k: released model configuration |
| License | Apache License 2.0.Plume 128K: official model card |
| Model size | Gemma 2B architecture; total size varies with the vocabulary.Plume 128K: official model card |
| Architecture configuration | 18 decoder layers, 2,048 hidden width and 128,000 vocabulary entries.Plume128k: released model configuration |
Related models
Plume 256K is BSC's translation research model with a 256,000-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