Aye10032/Qwen3-ASR-Refiner-0.6B

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Aye10032/Qwen3-ASR-Refiner-0.6B is a 0.8 billion parameter model from the Qwen3 family, fine-tuned to convert Chinese ASR transcripts and spoken-style text into formal, natural written Chinese. It preserves original meaning without adding new information, making it specialized for refining speech-to-text outputs. This model is part of a family trained on the Aye10032/WenetSpeech-Formal-Text dataset.

Loading preview...

Overview

Aye10032/Qwen3-ASR-Refiner-0.6B is a specialized model within the Qwen3 family, developed by Aye10032. It focuses on refining Chinese Automatic Speech Recognition (ASR) transcripts and other spoken-style text into concise, natural, and formal written Chinese. The model ensures that the original meaning is preserved and no new information is introduced during the refinement process.

Key Capabilities

  • Chinese ASR Transcript Refinement: Transforms raw ASR outputs into polished written Chinese.
  • Spoken-to-Written Conversion: Converts informal spoken language into formal, natural written text.
  • Meaning Preservation: Designed to maintain the original intent and information content of the input.
  • Family of Models: This 0.6B parameter variant is part of a larger family, including 1.7B and 4B parameter versions, all fine-tuned with the same methodology.

Training Details

The model was fine-tuned on the Aye10032/WenetSpeech-Formal-Text dataset, which provides paired examples of spoken and formal written Chinese. The LoRA adapter used during fine-tuning has been merged into the base model, allowing for direct loading of complete BF16 Transformers weights without requiring PEFT.