Downloads · 30 days
0
jashansinghTT/minimax-music3-blackhole
minimax-music3-blackhole is a text-to-audio model from jashansinghTT. Use it for the text-to-audio task on the model card, and read the license before you ship it in a product. The card lists the license as other.
Derived from MiniMaxAI/MiniMax-Music3 by MiniMax, unmodified. See the original model card for training and evaluation details.
Downloads · 30 days
0
Access
Public
Updated Oct 19, 2026
Repo size
3.5 GB
Likes
0
Public
Click a slice to open those files.
Other904 MB · 99%
From the Hugging Face model README
Derived from MiniMaxAI/MiniMax-Music3 by MiniMax, unmodified. See the original model card for training and evaluation details.
MiniMax Music 3, a lyrics-and-caption conditioned music generation model (8B Qwen3-based global LLM, 0.6B depth decoder for the residual RVQ codebooks, 2.4B flow-matching DiT and a DAC-style vocoder), with an SGLang-Omni compatible /v1/audio/speech API plus an async jobs API for multi-minute songs.
Runs on Tenstorrent p150 (mesh P150).
Packaged and published with tt-model-manager 0.1.0 (manifest schema 5.1).
| Hardware | p150 |
| License | minimax-music3-community-license |
Generating songs from lyrics and a text description of the music.
tt: uv tool install tenstorrent (or use tt-model alone, see the Quickstart)tt serve jashansinghTT/minimax-music3-blackhole
tt serve downloads the Docker image and the MiniMaxAI/MiniMax-Music3 weights at fbdf52fbaaca799592917417eb05f1899f1255ec (into your HF cache; they are not in the image), then starts the model's own HTTP server on port 20000 (or the next free port, if that one is busy). The first start compiles kernels for your device, which takes several minutes; the server is ready when it logs Application startup complete.
Without tt-cli: tt-model serve jashansinghTT/minimax-music3-blackhole.
curl -s localhost:20000/v1/audio/speech -H 'Content-Type: application/json' -d '{
"model": "MiniMaxAI/MiniMax-Music3",
"input": "[Verse]\nMorning light filtering through the pine\n[Chorus]\nSoftly the world begins to breathe",
"instructions": "A warm acoustic pop song with intimate female vocals, fingerpicked guitar and soft piano.",
"response_format": "wav", "seed": 7, "max_new_tokens": 750}' -o song.wav
This revision fixes first starts that could hang in ttnn.load_tensor at 100 % system time: on some hosts, reading a tensor-cache file soon after it was written hangs. The KV caches start as zeros and are now built without a cache file, and each start first copies the cache files the previous start wrote onto new files, once (a copy loads normally). Measured on a QuietBox 2 from empty caches: first start ready in 171 s, second start 90 s (including the one-time copy of about 21 GB), third 50 s. Output is unchanged: a seeded clip is byte-identical to the previous revision's.
This package exposes the model's own HTTP server on port 20000, not an OpenAI-compatible chat API — its request and response shapes are the model's own. See the author's notes for the payload it expects.
max_new_tokens counts audio frames (25 per second). Long songs: POST /v1/music/jobs,
poll GET /v1/music/jobs/{id}, then fetch GET /v1/music/jobs/{id}/audio.
| profile | request | audio length | concurrency | N | wall time |
|---|---|---|---|---|---|
p150 | POST /v1/audio/speech, max_new_tokens 250, WAV | 10 s | 1 | 1 | 44.7 s |
Measured end to end through this package's server in the tt-model community sweep, September 2026. No quality evaluation against the reference implementation has been run yet.
p150 is validated.Model weights are under the MiniMax-Music3 Community License; products built with it must display "MiniMax-Music3".
Questions or problems with this package: open a discussion at https://huggingface.co/jashansinghTT/minimax-music3-blackhole/discussions — that is what reaches its author. A problem with the tt tooling itself: tt report issue (collects your environment and opens a prefilled issue against tenstorrent/tt-cli). Product feedback: [email protected].
The exact sources the image was built from — code/ in this repo is byte-identical to the model code inside the image:
| component | built from |
|---|---|
| tt-metal | a local checkout — commit not published |
code/ digest | b131ad02856e6ea7 (sha256, first 16 hex digits) |
| built | 2026-09-12T09:31:21+00:00 by tt-model 0.1.0 |