ynklab/Tower-7B-Stair_FS4

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

Tower-7B-Stair_FS4 is a 7 billion parameter language model developed by ynklab, fine-tuned from Unbabel/TowerInstruct-Mistral-7B-v0.2. This model specializes in multilingual chunk-level machine translation, supporting English to and from German, Spanish, French, Italian, Korean, Dutch, Portuguese, Russian, and Chinese. It employs a unique fixed-range chunking and stair-step context scheme for document-level translation, making it particularly effective for maintaining length consistency across translated documents.

Loading preview...

Model Overview

Tower-7B-Stair_FS4 is a 7 billion parameter model developed by ynklab, derived from Unbabel/TowerInstruct-Mistral-7B-v0.2. It is specifically fine-tuned for multilingual chunk-level machine translation, addressing the challenge of length consistency in document-level translation.

Key Features and Training

  • Document-Level Translation: Utilizes a novel Fixed-Range Chunking (FRC) method, segmenting documents into 256–512-token ranges and merging them into four chunks for training.
  • Stair-Step Context Scheme: Employs a unique context mechanism where each subsequent chunk in a document is trained with increasing preceding context (e.g., the fourth chunk uses the first three chunks as context).
  • Multilingual Support: Capable of translating between English and nine other languages in both directions: German, Spanish, French, Italian, Korean, Dutch, Portuguese, Russian, and Chinese.
  • Prompt Formats: Fine-tuned with specific ChatML-style prompt formats (0c1t, 1c1t, 2c1t, 3c1t) tailored for different numbers of context chunks.
  • Training Details: Full-parameter supervised fine-tuning over 2 epochs, with a maximum sequence length of 32,768 tokens, using bfloat16 precision.

Use Cases

This model is ideal for developers and researchers focused on:

  • Document-level machine translation where maintaining context and length consistency across chunks is critical.
  • Translating between English and the supported languages, particularly for structured or long-form content.
  • Applications requiring context-aware translation of text segments within a larger document.

Finetunes
1 models