Rumble-2-Flash
Run Rumble-2-Flash Speech Recognition Locally
Deploy Reson8's 0.5B-parameter Rumble-2-Flash model locally via Hugging Face Transformers with custom formatting tokens.
- Intermediate
- 3 steps
- Free (Apache 2.0)
By TWK Guides DeskUpdated 19 Nov 2026
Every step checked against Reson8's own docs on 18 Nov 2026

Overview
Rumble-2-Flash is an open 0.5B-parameter speech recognition model built on Qwen3-ASR that supports 45 languages and lets you control digit formatting, filler words, and verbatim repeats.
Before you start
- Python 3.10 or newer with PyTorch and Hugging Face Transformers installed
- A local GPU or Apple Silicon Mac with at least 4 GB of memory
Step by step
0 of 3 done
Your ticks stay in this browser.
Step 1 of 3
Inspect the model card and licence
Confirm the Apache 2.0 licence and supported formatting tokens before downloading weights.
Do this
0 of 2 done
What you should see
You have confirmed the model ID Reson8/Rumble-2-Flash and its Apache 2.0 licence.
If it didn't work
Update huggingface_hub and transformers to the latest release if the architecture is not recognised.
Checked againsthuggingface.co
Frequently asked questions
Can I serve Rumble-2-Flash in production with vLLM?
Yes. Reson8 documents both Hugging Face Transformers and vLLM integration on the official model card.



