GGUF + pure-C++ runtime in CrispASR (companion to MiMo-V2.5-ASR)
#2
by cstr - opened
The MiMo-Audio-Tokenizer is the required companion to MiMo-V2.5-ASR in CrispASR — MiMo-ASR is token-based ASR (8-channel RVQ → text), so we need the tokenizer GGUF too.
We ship it at: cstr/mimo-tokenizer-GGUF
Loaded via --codec-model on the mimo-asr backend:
./build/bin/crispasr --backend mimo-asr \
-m mimo-asr-q4_k.gguf \
--codec-model mimo-tokenizer-q4_k.gguf \
-f audio.wav
The pair (LM + tokenizer) transcribes JFK verbatim against the upstream MimoAudio.asr_sft reference. C++ binary, GGUF — no Python, no torchaudio. Companion ASR repo: cstr/mimo-asr-GGUF.