ynklab/Tower-7B-Stair_2c1t

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Sep 7, 2026License:cc-by-nc-sa-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ynklab/Tower-7B-Stair_2c1t is a 7 billion parameter language model, fine-tuned from Unbabel/TowerInstruct-Mistral-7B-v0.2, specifically designed for multilingual chunk-level machine translation. It employs a unique stair-step context scheme, utilizing up to two preceding source-language chunks to enhance translation consistency. This model supports translation between English and nine other languages, making it suitable for document-level machine translation tasks where context preservation is crucial.

Loading preview...

Tower-7B-Stair_2c1t: Context-Aware Multilingual Machine Translation

Tower-7B-Stair_2c1t is a specialized 7 billion parameter model, derived from Unbabel/TowerInstruct-Mistral-7B-v0.2, focusing on document-level machine translation. Developed by ynklab, this model is a key component of the research presented in the paper "Doc2FRC: Length-Consistent Document-Level Machine Translation via Fixed-Range Chunking".

Key Capabilities and Features

  • Stair-Step Context Scheme: Uniquely handles document context by incorporating zero, one, or two preceding source-language chunks based on the current chunk's position. This helps maintain consistency across translated documents.
  • Multilingual Support: Facilitates translation between English and nine other languages, including German, Spanish, French, Italian, Korean, Dutch, Portuguese, Russian, and Chinese, in both directions.
  • Chunk-Level Translation: Optimized for translating text in fixed-range chunks (256–512 tokens), making it suitable for processing longer documents.
  • Fine-tuned for Consistency: The training methodology, using the Stair_2c1t variant and derived from the sardinelab/DocBlocks dataset, aims to produce more coherent document translations.

Recommended Use Cases

  • Document-Level Machine Translation: Ideal for translating entire documents where maintaining contextual flow and consistency across chunks is critical.
  • Research in Machine Translation: Useful for researchers exploring context-aware translation methods and chunking strategies.
  • Multilingual Content Processing: Applicable for tasks requiring translation between English and the supported languages, especially for structured or long-form content.

Finetunes
1 models