tzh699/GujinBridge-Qwen3-1.7B-DPO

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 7, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

GujinBridge-Qwen3-1.7B-DPO by tzh699 is a 2 billion parameter Qwen3-1.7B model specifically adapted for classical Chinese language tasks, including translation to modern Chinese, modern-to-classical rewriting, and classical Chinese punctuation. It leverages a DPO (Direct Preference Optimization) fine-tuning approach, building upon an SFT (Supervised Fine-Tuning) base. This model excels in specialized classical Chinese processing, making it ideal for applications requiring accurate interpretation and manipulation of ancient texts.

Loading preview...

GujinBridge-Qwen3-1.7B-DPO Overview

GujinBridge-Qwen3-1.7B-DPO is a 2 billion parameter language model developed by tzh699, built upon the Qwen3-1.7B architecture. It is specifically fine-tuned for tasks involving classical Chinese, including translation from classical to modern Chinese, rewriting modern Chinese into classical style, and classical Chinese punctuation. The model's development involved a multi-stage process: Qwen3-1.7B -> GujinBridge V5 LoRA SFT -> GujinBridge DPO V1, with the DPO checkpoint selected for its performance.

Key Capabilities

  • Classical Chinese Translation: Translates classical Chinese texts into modern Chinese.
  • Modern-to-Classical Rewriting: Converts modern Chinese into classical Chinese style.
  • Classical Chinese Punctuation: Accurately punctuates classical Chinese passages.
  • Specialized Fine-tuning: Utilizes Direct Preference Optimization (DPO) for enhanced performance in its target domains.

Good For

  • Human-assisted Translation: Generating candidate translations and punctuations for classical Chinese.
  • Educational Applications: Supporting learning and experimentation with classical Chinese texts.
  • Research and Development: Exploring SFT/DPO/GRPO pipelines and demonstrating classical Chinese processing capabilities.
  • Textual Analysis: Aiding in the interpretation and manipulation of ancient Chinese literature, with independent verification against primary sources recommended due to inherent limitations.

Finetunes
1 models