PraxySante/Qwen3-0.6B-SFT-ASR-Correction-FR-v6

TEXT GENERATIONConcurrency Cost:1Model Size:0.8BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 1, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

PraxySante/Qwen3-0.6B-SFT-ASR-Correction-FR-v6 is a 0.8 billion parameter Qwen3-based model developed by PraxySante, fine-tuned for French Automatic Speech Recognition (ASR) error correction specifically within the medical domain. This model specializes in refining transcribed medical text, leveraging its instruction-tuned architecture to improve accuracy. It is designed to process and correct French medical transcriptions, making it suitable for applications requiring high-fidelity text from speech in healthcare contexts.

Loading preview...

Overview

PraxySante/Qwen3-0.6B-SFT-ASR-Correction-FR-v6 is a specialized Qwen3-0.6B model, fine-tuned by PraxySante for correcting French Automatic Speech Recognition (ASR) errors, particularly in the medical field. With 0.8 billion parameters and a context length of 32768 tokens, this model is engineered to enhance the accuracy of medical transcriptions.

Key Capabilities

  • French ASR Correction: Specifically trained to identify and rectify errors in French speech-to-text outputs.
  • Medical Domain Specialization: Optimized for terminology and context relevant to healthcare, improving correction quality for medical dictations and records.
  • Instruction-Tuned: Utilizes a chat template with a system role as a "medical transcription correction assistant" to guide its responses.
  • Input Preprocessing: Recommends normalization steps for numbers, medications, units, and abbreviations to improve correction performance.

Training Details

The model was fine-tuned using Supervised Fine-Tuning (SFT) with LoRA on the PraxySante/Qwen3-0.6B-SFT-ASR-Correction-FR-vf-train-clean dataset. Training involved 35,000 steps, achieving a loss of approximately 0.70.

Good For

  • Developers building applications that require highly accurate French medical transcriptions.
  • Healthcare systems needing to refine ASR outputs for clinical documentation, patient records, or research.
  • Use cases where precise correction of medical terminology, numbers, and units in French is critical.