0

Rumble-2-Flash: an open 0.5B speech model for 40+ languages

https://huggingface.co/blog/rumble-2-flash(huggingface.co)
Rumble-2-Flash is a new open-source 0.5B parameter speech recognition model that supports over 40 languages. It was developed using on-policy distillation from a larger model, a process that allows it to achieve performance competitive with models up to 60 times its size. The model demonstrates strong accuracy with a 6.96% average error rate on the Open ASR Leaderboard and performs well on real-world phone call data, comparable to much larger proprietary models. Built on the Qwen3-ASR architecture, it is extremely fast and integrates directly with standard tools like 🤗 Transformers and vLLM without custom code.
0 points•by ogg•2 hours ago

Comments (0)

No comments yet. Be the first to comment!

Have an account? Log in to join the discussion.