0.8B local fine-tune matches frontier model on dictation cleanup
SpeakoFlow Mini, an Apache-2.0 fine-tune of Qwen3.5-0.8B, matches a hosted frontier model on the narrow task of dictation cleanup. It applies corrections the speaker made and leaves everything else unchanged.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- OpenAI product lead Tara Seshan discusses persistent AI coworkers
- DLSS 5 neural rendering: 150 MB model, real-time at 40% compute
- Krea 3 rumored to add editing, may open weights
- How LLMs Actually Work: A Walkthrough of Transformer Architecture
- MiniMax H3 image editor workflow generates character sheets locally