Running Voxtral-Mini-3B-2507 on GPU requires ~9.5 GB of GPU RAM in bf16 or fp16.
Running Voxtral-Small-24B-2507 on GPU requires ~55 GB of GPU RAM in bf16 or fp16.
kamranjon · 1h ago
Im pretty excited to play around with this. I’ve worked with whisper quite a bit, it’s awesome to have another model in the same class and from Mistral, who tend to be very open. I’m sure unsloth is already working on some GGUF quants - will probably spin it up tomorrow and try it on some audio.
ipsum2 · 6h ago
24B is crazy expensive for speech transcription. Conspicuously no comparison with Parakeet, a 600M param model thats currently dominating leaderboards (but only for English)
azinman2 · 17m ago
But it also includes world knowledge, can do tool calls, etc. It’s an omnimodel
sheerun · 2h ago
In demo they mention polish prononcuation is pretty bad, spoken as if second language of english-native speaker. I wonder if it's the same for other languages. On the other hand whispering-english is hillariously good, especially different emotions.
Does it support realtime transcription? What is the ~latency?
danelski · 13h ago
They claim to undercut competitors of similar quality by half for both models, yet they released both as Apache 2.0 instead of following smaller - open, larger - closed strategy used for their last releases.
What's different here?
halJordan · 6h ago
They didn't release voxtral large so your question doesn't really make sense
wmf · 7h ago
They're working on a bunch of features so maybe those will be closed. I guess they're feeling generous on the base model.
Havoc · 7h ago
Probably not looking to directly compete in transcription space
Running Voxtral-Small-24B-2507 on GPU requires ~55 GB of GPU RAM in bf16 or fp16.
Running Voxtral-Small-24B-2507 on GPU requires ~55 GB of GPU RAM in bf16 or fp16.