audio-cpp/MiniMax-Music3-GGUF

MiniMax Music 3 GGUF

GGUF package for MiniMax Music 3 for audio.cpp.

Upstream model: https://huggingface.co/MiniMaxAI/MiniMax-Music3 Upstream license: https://huggingface.co/MiniMaxAI/MiniMax-Music3/blob/main/LICENSE

Compatible Q8 GGUF package: https://huggingface.co/joemattie/MiniMax-Music3-GGUF

Preview Notes

• The implementation is available on the preview/minimax-music-3 branch while the runtime surface is iterated quickly.
• The current runtime uses model-local resource loading instead of treating the v1 spec as the runtime contract. This keeps component selection flexible while the package layout and option surface settle.
• The default component mix favors Q4_0 for the large language model and flow transformer, with Q8_0 for the RVQ depth decoder.
• BF16, Q8_0, and Q4_0 component variants are included for quality/performance comparison.
• Longer generations such as five-minute songs are supported as long-form runs, but they are currently tuned for completion and quality checks rather than realtime throughput.
• Memory usage remains an active optimization target for larger durations and alternate component mixes.

Performance Snapshot

Measured on an RTX 5090 with CUDA using a 30-second lyric generation request, 30 flow steps, flow guidance scale 1.7, AR guidance scale 1.5, and top-k 50. Peak VRAM is the observed nvidia-smi process peak during a warmup-plus-measured-request run.

Component mix	Language model	RVQ depth decoder	Flow transformer	RTF	Speed	Peak VRAM
Default Q4/Q8/Q4	q4_0	q8_0	q4_0	0.738	1.35x realtime	9.8 GiB
Q8	q8_0	q8_0	q8_0	0.832	1.20x realtime	13.4 GiB
BF16	bf16	bf16	bf16	1.389	0.72x realtime	19.4 GiB

Quick Start

audiocpp_cli \
  --task gen \
  --family minimax_music3 \
  --model MiniMax-Music3-GGUF \
  --backend cuda \
  --text "A bright pop rock song with clean drums and a clear male vocal." \
  --request-option 'lyrics=[verse] City lights are shining low. I keep moving with the glow. [chorus] Turn it up and let it fly. Sing the melody tonight.' \
  --request-option duration_sec=20 \
  --request-option num_inference_steps=30 \
  --out output.wav

Components

Default audio.cpp component mix:

• language_model_q4_0.gguf
• rvq_depth_decoder_q8_0.gguf
• transformer_q4_0.gguf
• condition_encoder.gguf
• vocoder.gguf

The BF16, Q8_0, and Q4_0 component variants are included for measurement and quality/performance comparison.

Component GGUFs can be selected explicitly for experiments:

--session-option minimax_music3.language_model_gguf=language_model_bf16.gguf
--session-option minimax_music3.language_model_gguf=language_model_q8_0.gguf
--session-option minimax_music3.rvq_depth_decoder_gguf=rvq_depth_decoder_bf16.gguf
--session-option minimax_music3.rvq_depth_decoder_gguf=rvq_depth_decoder_q8_0.gguf
--session-option minimax_music3.flow_transformer_gguf=transformer_bf16.gguf
--session-option minimax_music3.flow_transformer_gguf=transformer_q8_0.gguf

License

This GGUF package follows the upstream MiniMax-Music3 COMMUNITY LICENSE. The license text is included in LICENSE; review it before use, especially for commercial deployment.