cstr/moonshine-base-GGUF

Moonshine Base -- GGUF

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

Available variants

File	Quant	Size	Notes
moonshine-base.gguf	F32	235 MB	Full precision
moonshine-base-q4_k.gguf	Q4_K	40 MB	Best size/quality tradeoff

Model details

• Architecture: Conv1d stem + 8L transformer encoder + 8L transformer decoder (416d, 8 heads, partial RoPE, SiLU/GELU)
• Parameters: 61M
• Languages: English only
• License: MIT
• Source: UsefulSensors/moonshine-base

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-base-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
• Base models use head_dim=52 which works on CPU flash_attn

Provenance and EU AI Act Art. 53 note

• Upstream model: UsefulSensors/moonshine-base — published by UsefulSensors.
• Upstream licence: mit. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF/GGML). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.