Add MiniMax Music3 GGUF package

.gitattributes

@@ -33,3 +33,12 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+condition_encoder.gguf filter=lfs diff=lfs merge=lfs -text
+language_model_bf16.gguf filter=lfs diff=lfs merge=lfs -text
+language_model_q4_k.gguf filter=lfs diff=lfs merge=lfs -text
+rvq_depth_decoder_bf16.gguf filter=lfs diff=lfs merge=lfs -text
+rvq_depth_decoder_q4_k.gguf filter=lfs diff=lfs merge=lfs -text
+tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+transformer_bf16.gguf filter=lfs diff=lfs merge=lfs -text
+transformer_q4_k.gguf filter=lfs diff=lfs merge=lfs -text
+vocoder.gguf filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,127 @@
+MiniMax-Music3 COMMUNITY LICENSE
+
+Copyright (c) 2026 MiniMax
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this Software,
+including the model weights, parameters, configuration files, inference code and associated
+documentation (the “Software”), to deal in the Software, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or provide copies of the Software,
+and to permit persons to whom the Software is furnished to do so, subject to the following
+conditions:
+
+1. The above copyright notice and this permission notice shall be included in all copies or
+substantial portions of the Software.
+
+2. Your use of the Software must comply with applicable laws and regulations (including
+trade-compliance laws and regulations) and must comply with the Acceptable Use Policy in Exhibit A.
+Especially, you must not violate any third party’s intellectual-property or other rights while using
+the Software.
+
+3. Commercial Terms
+
+    1. You shall prominently display “MiniMax-Music3” on the user interface of commercial product or
+    service that uses the Software.
+
+    2. You shall obtain a separate, prior written authorization from MiniMax by contacting
+    api@minimax.io with the subject line “MiniMax-Music3 licensing - authorization request”, if the
+    aggregate yearly revenue generated from such products and services provided by you and/or any of
+    your affiliates exceeds 20 million US dollars (or equivalent in other currencies).
+
+4. If you provide or make available to any third party a product, service, or hosted service that
+permits the generation of outputs using the Software, you must, before making that product or
+service available and throughout its operation, implement, maintain, test, and periodically review
+reasonable and proportionate technical and organizational safeguards designed to prevent and
+mitigate access, uses, and outputs that violate this License, including uses or outputs that
+infringe, misappropriate, or otherwise violate any third party’s intellectual-property or other
+rights. You must not knowingly disable, materially weaken, or permit the circumvention of those
+safeguards. You are responsible for implementing and enforcing these requirements with respect to
+your products, services, systems, users, and downstream recipients.
+
+THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES
+OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. YOU SHALL BE SOLELY
+RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING, REPRODUCING, MODIFYING, PERFORMING,
+DISPLAYING OR DISTRIBUTING ANY OF THE SOFTWARE OR OUTPUTS AND ASSUME ANY AND ALL RISKS ASSOCIATED
+WITH YOUR OR A THIRD PARTY’S USE OR DISTRIBUTION OF ANY OF THE SOFTWARE OR OUTPUTS AND YOUR EXERCISE
+OF RIGHTS AND PERMISSIONS UNDER THIS LICENSE. YOU SHALL INDEMNIFY AND HOLD HARMLESS MINIMAX FROM AND
+AGAINST ANY AND ALL LOSSES, LIABILITIES, DAMAGES, CLAIMS, COSTS, AND EXPENSES ARISING OUT OF OR
+RELATING TO ANY OF YOUR BREACH OF THIS LICENSE.
+
+Please note that MiniMax-Music3 was fine-tuned from Qwen3-8B (https://huggingface.co/Qwen/Qwen3-8B),
+which is licensed under the Apache License 2.0
+(https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE). DiT-2B was modified from the  Stable Audio
+code (https://github.com/Stability-AI/stable-audio-tools), which is licensed under the MIT License
+(https://github.com/Stability-AI/stable-audio-tools/blob/main/LICENSE). The VAE was modified from
+the DAC code (https://github.com/descriptinc/descript-audio-codec), which is licensed under the MIT
+License (https://github.com/descriptinc/descript-audio-codec/blob/main/LICENSE).
+
+Exhibit A — Acceptable Use Policy
+
+MiniMax reserves the right to update this Acceptable Use Policy from time to time.
+
+Last revised: August [6], 2026.
+
+MiniMax is committed to promoting the safe and responsible use of its tools and features. You agree
+not to use the Software and in any of the following ways:
+
+1. Use in any manner that violates any applicable national, federal, state, local, or international
+law, regulation, or other legal requirement, or that infringes, misappropriates, or otherwise
+violates any third party’s intellectual-property or other proprietary rights, including without
+limitation, through unauthorized reproduction, distribution, public display, public performance, or
+creation of derivative works;
+
+2. Use in any manner that may harm yourself or others;
+
+3. Use to generate, repurpose or distribute any content to harm yourself or others;
+
+4. Use to circumvent or bypass any safety guardrails or safeguards we have implemented;
+
+5. Use in any manner that exploits or harms, or intends to exploit or harm, minors;
+
+6. Use to generate or disseminate verifiably false information and/or content for the purpose of
+harming others or influencing elections;
+
+7. Use to manufacture or facilitate false online engagement, including fake reviews and other means
+of false online engagement;
+
+8. Use to intentionally defame, disparage, or otherwise harass others;
+
+9. Use to generate and/or disseminate malware (including ransomware) or any other content intended
+to damage electronic systems;
+
+10. Use to generate or disseminate personally identifiable information for the purpose of harming
+others;
+
+11. Use to generate or disseminate information (including images, code, posts, or articles) in or to
+any public environment (including via bot tweets or similar means) without clearly and prominently
+disclosing that such information and/or content is machine-generated;
+
+12. Use to impersonate another person without that person’s consent, authorization, or lawful right
+to do so;
+
+13. Use to make high-risk automated decisions in critical domains that affect individual safety,
+rights, or well-being (such as law enforcement, immigration, healthcare or medical services,
+critical-infrastructure management, product-safety components, essential services, credit,
+employment, housing, education, social scoring, or insurance);
+
+14. Use in any manner that violates or disregards the social, ethical, or moral standards of other
+countries or regions;
+
+15. Use to carry out, assist, threaten, incite, plan, advocate for, or encourage violent extremism
+or terrorism;
+
+16. Use for any purpose intended to discriminate against, or harm, individuals or groups based on
+protected characteristics or categories, online or offline social behavior, or known or predicted
+personality traits;
+
+17. Use to intentionally exploit the vulnerabilities of specific populations based on age, social,
+physical, or psychological characteristics, so as to materially distort the behavior of a member of
+that group in a manner that causes, or is likely to cause, physical or psychological harm to that
+person or to others;
+
+18. Use for military purposes; and
+
+19. Use to engage in any unauthorized or unlicensed professional activity, including but not limited
+to financial, legal, medical or healthcare, or other professional practice.

README.md

@@ -0,0 +1,70 @@
+---
+library_name: audio.cpp
+pipeline_tag: text-to-audio
+license: other
+license_name: minimax-music3-community-license
+base_model:
+  - MiniMaxAI/MiniMax-Music3
+tags:
+  - audio.cpp
+  - gguf
+  - minimax-music3
+---
+
+# 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
+
+The audio.cpp implementation is being staged on the `preview/minimax-music-3`
+branch while the runtime surface is iterated quickly.
+
+## Quick Start
+
+```bash
+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_k.gguf`
+- `rvq_depth_decoder_bf16.gguf`
+- `transformer_q4_k.gguf`
+- `condition_encoder.gguf`
+- `vocoder.gguf`
+
+The BF16 and Q4_K component variants are included for measurement and quality/performance comparison.
+
+Component GGUFs can be selected explicitly for experiments:
+
+```bash
+--session-option minimax_music3.language_model_gguf=language_model_bf16.gguf
+--session-option minimax_music3.rvq_depth_decoder_gguf=rvq_depth_decoder_q4_k.gguf
+--session-option minimax_music3.flow_transformer_gguf=transformer_bf16.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.
+
+## Preview Notes
+
+- 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 best quality/performance component combination is still being measured. The default favors Q4_K for the large language model and flow transformer, with BF16 kept for the RVQ depth decoder.
+- 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 is an active optimization target for larger durations and alternate component mixes.

condition_encoder.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f9414097b1f6ad154dae08ed69ce88fd0db6d7fd422f4d1a847216ca28ba1bf
+size 100677184

config.json

@@ -0,0 +1,6 @@
+{
+  "architectures": [
+    "MiniMaxMusic3ForConditionalGeneration"
+  ],
+  "model_type": "minimax_music3"
+}

config/condition_encoder.json

@@ -0,0 +1,11 @@
+{
+  "_class_name": "MiniMaxMusic3ConditionEncoder",
+  "_diffusers_version": "0.40.0.dev0",
+  "condition_hidden_dim": 4096,
+  "input_hop_length": 960,
+  "input_sampling_rate": 24000,
+  "num_condition_layers": 8,
+  "out_dim": 2048,
+  "output_hop_length": 512,
+  "output_sampling_rate": 44100
+}

config/language_model.json

@@ -0,0 +1,71 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": null,
+  "dtype": "bfloat16",
+  "eos_token_id": null,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 12288,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 10240,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 8,
+  "pad_token_id": null,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 1000000,
+    "rope_type": "default"
+  },
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "5.13.0.dev0",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 200000
+}

config/rvq_depth_decoder.json

@@ -0,0 +1,11 @@
+{
+  "_class_name": "MiniMaxMusic3RVQDepthDecoder",
+  "_diffusers_version": "0.40.0.dev0",
+  "audio_vocab_size": 1024,
+  "hidden_size": 4096,
+  "intermediate_size": 6144,
+  "max_position_embeddings": 16,
+  "num_attention_heads": 16,
+  "num_codebooks": 8,
+  "num_layers": 4
+}

config/transformer.json

@@ -0,0 +1,12 @@
+{
+  "_class_name": "MiniMaxMusic3Transformer1DModel",
+  "_diffusers_version": "0.40.0.dev0",
+  "attention_head_dim": 64,
+  "condition_dim": 2048,
+  "ff_inner_dim": 8192,
+  "fourier_embedding_dim": 256,
+  "in_channels": 128,
+  "num_attention_heads": 32,
+  "num_layers": 36,
+  "rotary_dim": 32
+}

config/vocoder.json

@@ -0,0 +1,14 @@
+{
+  "_class_name": "MiniMaxMusic3Vocoder",
+  "_diffusers_version": "0.40.0.dev0",
+  "decoder_hidden_dim": 1536,
+  "decoder_input_dim": 1024,
+  "latent_channels": 128,
+  "sampling_rate": 44100,
+  "upsampling_ratios": [
+    8,
+    8,
+    4,
+    2
+  ]
+}

language_model_bf16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76ccb207d9ec49a53558f242a084087f59242abfc7805db46e3e8b4454e1f237
+size 17169011872

language_model_q4_k.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:019300ae45ae28b193b3deb2ec519c7ae8e255abe90a29b39e2a7044166e05ce
+size 7184471360

rvq_depth_decoder_bf16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f7adfb54a9e156682555dc7ddc575c20a6f038c8d036e364881a3ea80400fd0
+size 1292061152

rvq_depth_decoder_q4_k.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c5d41b27418d9c1046345f649cb61d7cde0e3bbda4af7f7cb142df2c70cbdd0
+size 405752480

tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1537fa9e59a537276ecbc2e12d0438edff635a1f4a4948e679774b5feb3e610
+size 11423801

tokenizer/tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

transformer_bf16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f855f1f285fccf9fdcb0b88d0d3514cbed1d884dc5ea290d4745886888159f5
+size 9727686016

transformer_q4_k.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ab8a79d251d05a128964bb0e104fdacbed7199dfe9e5b9e5e1e5b0605d2e5b2
+size 1396397600

vocoder.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a768b596516678fc850da1aff36556b8bc6bafcb7b42bc7ba7cb46cb9927a42
+size 216704192