R2T2 streaming ASR: frozen audio encoder, decoder-only tuning with SFT then GRPO
dr_cintas · x · 2026-11-06
drcintas summarizes the training recipe behind R2T2, a streaming ASR approach built on Qwen3-ASR:
- Post-trained on 1,500h of Chinese and English audio; the audio encoder and projector stay frozen, only the language decoder is updated, with no architectural changes.
- Two-stage pipeline: SFT followed by GRPO reinforcement learning.
- The insight is in the objective, not the architecture — so it works with vLLM/Transformers out of the box.
A lightweight, reproducible recipe for turning a full-context ASR model into a streaming one by touching only the language side.
More from Research
- Unverified claims: frontier labs allegedly gaming ICLR reviews with malicious reviews — examachine · 2026-11-07
- SSF teaches robots to reason about likely hiding spots when searching for hidden objects — _krishna_murthy · 2026-11-07
- 'Die getting Schmidhubered or live long enough to become one': scaling law priority drama — teortaxesTex · 2026-11-07
- Stanford and Together AI propose Next-Concept Modeling to replace next-token prediction — bendee983 · 2026-11-07
- New benchmark shows some LLMs refuse over 50% of legitimate biology queries — nlarusstone · 2026-11-07
- From AI-designed phage DNA to patient care: the ecosystem behind the model — exnx · 2026-11-07