PraxySante/qwen3-0.6b-sft-asr-correction-v15-context-full
PraxySante/qwen3-0.6b-sft-asr-correction-v15-context-full is a 0.8 billion parameter Qwen3-based model fine-tuned by PraxySante for medical ASR error correction in French. It specializes in correcting erroneous input by leveraging contextual information, making it suitable for improving the accuracy of speech-to-text transcripts in medical domains. The model was trained using Supervised Fine-Tuning on approximately 5.5 million paired examples, achieving a final evaluation loss of 0.073076.
Loading preview...
Model Overview
PraxySante/qwen3-0.6b-sft-asr-correction-v15-context-full is a specialized 0.8 billion parameter language model built upon the Qwen3-0.6B architecture. It has been specifically fine-tuned using Supervised Fine-Tuning (SFT) to address and correct Automatic Speech Recognition (ASR) errors within the medical domain, focusing exclusively on the French language.
Key Capabilities
- Context-Aware Correction: The model utilizes a contextual input structure, incorporating three fields:
context,input errone(erroneous input), andtarget corrige(corrected target), to enhance correction accuracy. - Medical ASR Error Remediation: Optimized for identifying and rectifying inaccuracies in French medical speech-to-text transcripts.
- Efficient Training: Trained on a substantial dataset of approximately 5.5 million paired examples, achieving a low final evaluation loss of 0.073076 over 79,836 steps.
Good For
- Improving ASR Accuracy: Ideal for applications requiring high accuracy in French medical transcription, where ASR systems frequently introduce errors.
- Specialized Language Tasks: Suited for use cases demanding precise linguistic correction within a specific, technical domain like medicine.
- French Language Processing: Particularly effective for tasks involving French text correction, especially when context is crucial for disambiguation.