cstr/moonshine-tiny-zh-GGUF

Moonshine Tiny (Chinese) -- GGUF

GGUF conversions and quantisations of UsefulSensors/moonshine-tiny-zh for use with CrispStrobe/CrispASR.

Available variants

File	Quant	Size	Notes
moonshine-tiny-zh.gguf	F32	104 MB	Full precision
moonshine-tiny-zh-q4_k.gguf	Q4_K	21 MB	Best size/quality tradeoff

Model details

• Architecture: Conv1d stem + 6L transformer encoder + 6L transformer decoder (288d, 8 heads, partial RoPE, SiLU/GELU)
• Parameters: 27M
• Languages: Chinese (fine-tuned from English moonshine-tiny)
• License: Moonshine AI Community License (free for <$1M revenue, attribution required)
• Source: UsefulSensors/moonshine-tiny-zh

Usage with CrispASR

# Auto-download (English tiny only)
./build/bin/crispasr --backend moonshine -m auto -f audio.wav

# Explicit model path
./build/bin/crispasr --backend moonshine -m moonshine-tiny-zh-q4_k.gguf -f audio.wav

Notes

• Moonshine models run on CPU only (GPU not needed for these small models)
• Tokenizer (tokenizer.bin) must be in the same directory as the model file
• Tiny models use head_dim=36 which works on CPU flash_attn