Whisper Large v3 Turbo
Pruned and fine-tuned Whisper Large v3 with 4 decoder layers instead of 32. OpenAI lists ~8x relative speed with a minimal degradation in accuracy.
Overview
Pruned and fine-tuned Whisper Large v3 with 4 decoder layers instead of 32. OpenAI lists ~8x relative speed with a minimal degradation in accuracy.
Family & lineage
How this model relates to others in its lineage. Family members share architecture and training-data roots; parent / children edges record direct distillation or fine-tune relationships.
Strengths
- MIT license
- ~8x relative speed vs large, per OpenAI's README (https://github.com/openai/whisper)
Weaknesses
- Slight accuracy drop on rare languages
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| FP16 | 1.6 GB | 2 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of Whisper Large v3 Turbo.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run Whisper Large v3 Turbo?
Can I use Whisper Large v3 Turbo commercially?
What's the context length of Whisper Large v3 Turbo?
Does Whisper Large v3 Turbo support images?
Source: huggingface.co/openai/whisper-large-v3-turbo
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify Whisper Large v3 Turbo runs on your specific hardware before committing money.