ynklab/Tower-7B-Uni_2c1t
The ynklab/Tower-7B-Uni_2c1t is a 7 billion parameter language model, fine-tuned from Unbabel/TowerInstruct-Mistral-7B-v0.2, specifically for multilingual chunk-level machine translation. It supports translation between English and nine other languages, leveraging up to two preceding source-language context chunks for improved length consistency. This model excels at document-level machine translation by processing text in fixed-range chunks of 256-512 tokens.
Loading preview...
Overview
Tower-7B-Uni_2c1t is a 7 billion parameter model developed by ynklab, fine-tuned from Unbabel/TowerInstruct-Mistral-7B-v0.2. Its primary purpose is multilingual chunk-level machine translation, specifically designed to maintain length consistency across document translations. The model was trained using a unique approach called Uni_2c1t, which jointly processes three prompt forms: no context, one preceding source-language context chunk, or two preceding source-language context chunks.
Key Capabilities
- Multilingual Translation: Supports translation between English and German, Spanish, French, Italian, Korean, Dutch, Portuguese, Russian, and Chinese in both directions.
- Context-Aware Translation: Utilizes up to two preceding source-language context chunks to enhance translation quality and consistency, particularly for document-level tasks.
- Fixed-Range Chunking: Optimized for processing text segmented into fixed-range chunks (256–512 tokens), derived from the sardinelab/DocBlocks dataset.
- Prompt Flexibility: Fine-tuned to handle different prompt formats for zero, one, or two context chunks, allowing adaptable usage based on the available context.
Good For
- Document-Level Machine Translation: Ideal for scenarios requiring consistent translation of longer documents by breaking them into manageable, context-aware chunks.
- Research in Contextual MT: Useful for researchers exploring the impact of source-language context on machine translation quality and length consistency.
- Multilingual Applications: Suitable for applications needing robust translation capabilities across the specified language pairs, especially where contextual understanding is critical.