ynklab/Qwen2.5-7B-Sep_1c1t
ynklab/Qwen2.5-7B-Sep_1c1t is a 7.6 billion parameter language model, fine-tuned from Qwen/Qwen2.5-7B-Instruct, specifically designed for multilingual chunk-level machine translation. Developed by ynklab, this model excels at translating between English and nine other languages, utilizing a fixed-range chunking approach with one preceding source-language context chunk. It supports a maximum sequence length of 32,768 tokens and is optimized for document-level translation tasks.
Loading preview...
Model Overview
ynklab/Qwen2.5-7B-Sep_1c1t is a 7.6 billion parameter model, fine-tuned from Qwen/Qwen2.5-7B-Instruct, specializing in multilingual chunk-level machine translation. This model is a Sep_1c1t variant, meaning it processes translations by considering one preceding source-language context chunk alongside the current source chunk, translating it into one target chunk. It was developed as part of the Doc2FRC: Length-Consistent Document-Level Machine Translation via Fixed-Range Chunking paper.
Key Capabilities
- Multilingual Translation: Supports translation between English and German, Spanish, French, Italian, Korean, Dutch, Portuguese, Russian, and Chinese in both directions.
- Context-Aware Chunking: Utilizes a fixed-range chunking method (256–512 tokens) with a preceding source-language context chunk to maintain document-level consistency.
- High Context Length: Features a maximum sequence length of 32,768 tokens, suitable for processing longer document segments.
- Fine-tuned Performance: Achieved through full-parameter supervised fine-tuning on the sardinelab/DocBlocks dataset over 2 epochs.
Good For
- Document-Level Machine Translation: Ideal for tasks requiring consistent translation across entire documents by leveraging chunk-level context.
- Research in MT: Useful for researchers exploring fixed-range chunking and context-aware translation methods, particularly those interested in the Doc2FRC paper's methodology.
- Specific Language Pairs: Excellent for applications needing translation between English and the nine supported languages, especially where context preservation is critical.