[Registry] MiniMax-Music3-condition_encoder-F32.gguf -> mm3-cond
[Registry] MiniMax-Music3-language_model-BF16.gguf -> mm3-lm
[Registry] MiniMax-Music3-language_model-Q5_K_M.gguf -> mm3-lm
[Registry] MiniMax-Music3-language_model-Q6_K.gguf -> mm3-lm
[Registry] MiniMax-Music3-language_model-Q8_0.gguf -> mm3-lm
[Registry] MiniMax-Music3-rvq_depth_decoder-BF16.gguf -> mm3-depth
[Registry] MiniMax-Music3-rvq_depth_decoder-Q8_0.gguf -> mm3-depth
[Registry] MiniMax-Music3-transformer-F32.gguf -> mm3-dit
[Registry] MiniMax-Music3-transformer-Q4_K_M.gguf -> mm3-dit
[Registry] MiniMax-Music3-transformer-Q5_K_M.gguf -> mm3-dit
[Registry] MiniMax-Music3-transformer-Q6_K.gguf -> mm3-dit
[Registry] MiniMax-Music3-transformer-Q8_0.gguf -> mm3-dit
[Registry] MiniMax-Music3-vocoder-F32.gguf -> mm3-vae
[Store] Created (policy=STRICT)
[Pipeline] Flash attention disabled
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
  Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/git/minimaxmusic.cpp/build/libggml-cuda.so
load_backend: loaded Vulkan backend from /mnt/workspace/git/minimaxmusic.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/git/minimaxmusic.cpp/build/libggml-cpu-zen4.so
[Load] LM backend: Vulkan0 (CPU threads: 16)
[GGUF] ../models/MiniMax-Music3-language_model-BF16.gguf: 399 tensors, data at offset 5346496
[LM-Config] 36L, H=4096, V=200000, Nh=32, Nkv=8, D=128, tied=0
[Qwen3] Attn: Q+K+V fused
[Qwen3] MLP: gate+up fused
[WeightCtx] Loaded 399 tensors, 16374.2 MB into backend
[LM-KV] Allocated 2 sets x 36 layers (4D batched), 2880.0 MB
[Store] Load LM: 2041 ms
[GGUF] ../models/MiniMax-Music3-rvq_depth_decoder-BF16.gguf: 47 tensors, data at offset 3456
[Load] Depth backend: Vulkan0 (shared)
[WeightCtx] Loaded 47 tensors, 1232.4 MB into backend
[Depth] Loaded: 4 layers, dim 4096, 7 heads
[Store] Load Depth: 144 ms
[BPE] Loaded from GGUF: 151675 vocab, 151387 merges
[Prompt] 40 tokens
[AR] Prefill 133 ms, 40 tokens, CFG=1.50, top_k=50, songs=1
[AR] Frame 0/200
[AR] Frame 100/200
[AR] Song 0: 200 frames (8.0s of music)
[Debug] frame_hiddens: [200, 8, 4096] first4: 0.921867 1.897966 -0.904081 -0.699011
[AR] 200 frames total, 29.7 s (148.6 ms/frame)
[Store] Unload Depth (1232.4 MB)
[Store] Unload LM (16374.2 MB)
[GGUF] ../models/MiniMax-Music3-condition_encoder-F32.gguf: 4 tensors, data at offset 640
load_backend: loaded CUDA backend from /mnt/workspace/git/minimaxmusic.cpp/build/libggml-cuda.so
load_backend: loaded Vulkan backend from /mnt/workspace/git/minimaxmusic.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/git/minimaxmusic.cpp/build/libggml-cpu-zen4.so
[Load] Cond backend: Vulkan0 (CPU threads: 16)
[Cond] Loaded: mix over 8 states, proj 4096 -> 2048
[Store] Load Cond: 48 ms
[GGUF] ../models/MiniMax-Music3-transformer-Q8_0.gguf: 441 tensors, data at offset 31712
[Load] DiT backend: Vulkan0 (shared)
[WeightCtx] Loaded 436 tensors, 2477.0 MB into backend
[DiT] Loaded: 36 layers, dim 2048, flash_attn=1
[Store] Load DiT: 333 ms
[GGUF] ../models/MiniMax-Music3-vocoder-F32.gguf: 182 tensors, data at offset 12896
[Load] VAE backend: Vulkan0 (shared)
[VAE] Backend: Vulkan0, Weight buffer: 103.4 MB
[VAE] Loaded: 4 blocks, upsample=512x, F32 activations
[Store] Load VAE: 106 ms
[Debug] noise: [689, 128] first4: 0.194019 2.161374 -0.172051 0.849060
[DiT] Graph: 1344 nodes, T=689, B=2
[Debug] temb_t: [2048] first4: 3.626598 -3.071267 -3.962117 4.059601
[Debug] hidden_after_preprocess: [689, 2304] first4: 0.058765 3.944272 -0.796806 2.207703
[Debug] hidden_after_proj_in: [689, 2048] first4: 3.507812 -5.736328 -1.914062 -0.175781
[Debug] layer0_sa_output: [690, 2048] first4: 1.452881 -1.077148 0.441460 1.429321
[Debug] hidden_after_layer0: [690, 2048] first4: -2.219409 -0.479538 -3.205898 3.676364
[Debug] hidden_after_layer6: [690, 2048] first4: -2.328953 2.079683 -2.211175 1.296423
[Debug] hidden_after_layer12: [690, 2048] first4: 0.123718 -0.422166 -0.041558 3.211436
[Debug] hidden_after_layer18: [690, 2048] first4: -2.070641 -2.252544 0.804997 0.110994
[Debug] hidden_after_layer35: [690, 2048] first4: 0.191328 1.356829 -2.184819 -1.811653
[Debug] dit_step0_vt_cond: [689, 128] first4: -0.121338 -2.163574 -0.451660 -0.892227
[Debug] dit_step0_vt_uncond: [689, 128] first4: -0.357056 -2.277603 -0.424774 -0.597168
[Debug] dit_step0_vt: [689, 128] first4: 0.043665 -2.083754 -0.470480 -1.098769
[Debug] dit_step0_xt: [689, 128] first4: 0.195474 2.091915 -0.187733 0.812434
[Debug] dit_step1_vt_cond: [689, 128] first4: -0.029053 -1.979492 -0.385742 -0.784180
[Debug] dit_step1_vt_uncond: [689, 128] first4: 0.266418 -2.145203 -0.856445 -1.080444
[Debug] dit_step1_vt: [689, 128] first4: -0.235883 -1.863495 -0.056250 -0.576794
[Debug] dit_step1_xt: [689, 128] first4: 0.187612 2.029799 -0.189608 0.793208
[Debug] dit_step2_vt_cond: [689, 128] first4: -0.000854 -1.700012 -0.423828 -0.583008
[Debug] dit_step2_vt_uncond: [689, 128] first4: -0.171997 -1.509521 -0.039062 -0.429260
[Debug] dit_step2_vt: [689, 128] first4: 0.118945 -1.833356 -0.693164 -0.690631
[Debug] dit_step2_xt: [689, 128] first4: 0.191576 1.968687 -0.212714 0.770187
[Debug] dit_step3_vt_cond: [689, 128] first4: 0.041748 -1.376953 -0.419434 -0.404785
[Debug] dit_step3_vt_uncond: [689, 128] first4: -0.289734 -1.246155 -0.147095 -0.270996
[Debug] dit_step3_vt: [689, 128] first4: 0.273785 -1.468512 -0.610071 -0.498438
[Debug] dit_step3_xt: [689, 128] first4: 0.200703 1.919737 -0.233049 0.753572
[Debug] dit_step4_vt_cond: [689, 128] first4: 0.084961 -1.077431 -0.406738 -0.248047
[Debug] dit_step4_vt_uncond: [689, 128] first4: -0.124023 -1.069946 -0.145386 -0.200195
[Debug] dit_step4_vt: [689, 128] first4: 0.231250 -1.082670 -0.589685 -0.281543
[Debug] dit_step4_xt: [689, 128] first4: 0.208411 1.883648 -0.252706 0.744188
[Debug] dit_step5_vt_cond: [689, 128] first4: 0.131104 -0.768673 -0.399902 -0.095215
[Debug] dit_step5_vt_uncond: [689, 128] first4: 0.099365 -0.855316 -0.248001 -0.118225
[Debug] dit_step5_vt: [689, 128] first4: 0.153320 -0.708023 -0.506233 -0.079108
[Debug] dit_step5_xt: [689, 128] first4: 0.213522 1.860047 -0.269580 0.741551
[Debug] dit_step6_vt_cond: [689, 128] first4: 0.184570 -0.440308 -0.403900 0.049316
[Debug] dit_step6_vt_uncond: [689, 128] first4: 0.299789 -0.573730 -0.269409 0.002472
[Debug] dit_step6_vt: [689, 128] first4: 0.103917 -0.346912 -0.498044 0.082108
[Debug] dit_step6_xt: [689, 128] first4: 0.216986 1.848483 -0.286182 0.744288
[Debug] dit_step7_vt_cond: [689, 128] first4: 0.236572 -0.079102 -0.411987 0.187317
[Debug] dit_step7_vt_uncond: [689, 128] first4: 0.520187 -0.122131 -0.292114 0.131622
[Debug] dit_step7_vt: [689, 128] first4: 0.038042 -0.048981 -0.495898 0.226303
[Debug] dit_step7_xt: [689, 128] first4: 0.218254 1.846850 -0.302711 0.751831
[Debug] dit_step8_vt_cond: [689, 128] first4: 0.283051 0.270508 -0.407227 0.311768
[Debug] dit_step8_vt_uncond: [689, 128] first4: 0.701172 0.382019 -0.274414 0.259155
[Debug] dit_step8_vt: [689, 128] first4: -0.009634 0.192450 -0.500195 0.348596
[Debug] dit_step8_xt: [689, 128] first4: 0.217932 1.853265 -0.319385 0.763451
[Debug] dit_step9_vt_cond: [689, 128] first4: 0.326935 0.627686 -0.407227 0.428955
[Debug] dit_step9_vt_uncond: [689, 128] first4: 0.781738 0.820068 -0.249756 0.376465
[Debug] dit_step9_vt: [689, 128] first4: 0.008572 0.493018 -0.517456 0.465698
[Debug] dit_step9_xt: [689, 128] first4: 0.218218 1.869699 -0.336633 0.778974
[Debug] dit_step10_vt_cond: [689, 128] first4: 0.362793 0.952637 -0.405273 0.546387
[Debug] dit_step10_vt_uncond: [689, 128] first4: 0.796875 1.178955 -0.238281 0.506226
[Debug] dit_step10_vt: [689, 128] first4: 0.058936 0.794214 -0.522168 0.574499
[Debug] dit_step10_xt: [689, 128] first4: 0.220183 1.896173 -0.354039 0.798124
[Debug] dit_step11_vt_cond: [689, 128] first4: 0.401367 1.260986 -0.407104 0.648560
[Debug] dit_step11_vt_uncond: [689, 128] first4: 0.789062 1.455688 -0.255790 0.633301
[Debug] dit_step11_vt: [689, 128] first4: 0.129980 1.124695 -0.513025 0.659241
[Debug] dit_step11_xt: [689, 128] first4: 0.224515 1.933663 -0.371140 0.820099
[Debug] dit_step12_vt_cond: [689, 128] first4: 0.457031 1.541504 -0.409424 0.753906
[Debug] dit_step12_vt_uncond: [689, 128] first4: 0.784180 1.684814 -0.289795 0.764160
[Debug] dit_step12_vt: [689, 128] first4: 0.228027 1.441187 -0.493164 0.746729
[Debug] dit_step12_xt: [689, 128] first4: 0.232116 1.981702 -0.387578 0.844990
[Debug] dit_step13_vt_cond: [689, 128] first4: 0.506836 1.784424 -0.426025 0.850098
[Debug] dit_step13_vt_uncond: [689, 128] first4: 0.791992 1.889893 -0.327637 0.866943
[Debug] dit_step13_vt: [689, 128] first4: 0.307227 1.710596 -0.494897 0.838306
[Debug] dit_step13_xt: [689, 128] first4: 0.242357 2.038722 -0.404075 0.872933
[Debug] dit_step14_vt_cond: [689, 128] first4: 0.559570 2.030273 -0.441406 0.928223
[Debug] dit_step14_vt_uncond: [689, 128] first4: 0.786133 2.067627 -0.374023 0.958984
[Debug] dit_step14_vt: [689, 128] first4: 0.400977 2.004126 -0.488574 0.906689
[Debug] dit_step14_xt: [689, 128] first4: 0.255723 2.105526 -0.420361 0.903156
[Debug] dit_step15_vt_cond: [689, 128] first4: 0.594727 2.215820 -0.458984 0.996094
[Debug] dit_step15_vt_uncond: [689, 128] first4: 0.780273 2.260986 -0.417480 1.031250
[Debug] dit_step15_vt: [689, 128] first4: 0.464844 2.184204 -0.488037 0.971484
[Debug] dit_step15_xt: [689, 128] first4: 0.271218 2.178333 -0.436629 0.935539
[Debug] dit_step16_vt_cond: [689, 128] first4: 0.621094 2.425781 -0.487793 1.044434
[Debug] dit_step16_vt_uncond: [689, 128] first4: 0.787109 2.415283 -0.451660 1.095215
[Debug] dit_step16_vt: [689, 128] first4: 0.504883 2.433130 -0.513086 1.008887
[Debug] dit_step16_xt: [689, 128] first4: 0.288047 2.259437 -0.453731 0.969169
[Debug] dit_step17_vt_cond: [689, 128] first4: 0.638672 2.543945 -0.506348 1.101074
[Debug] dit_step17_vt_uncond: [689, 128] first4: 0.780762 2.530762 -0.478027 1.144531
[Debug] dit_step17_vt: [689, 128] first4: 0.539209 2.553174 -0.526172 1.070654
[Debug] dit_step17_xt: [689, 128] first4: 0.306021 2.344543 -0.471271 1.004857
[Debug] dit_step18_vt_cond: [689, 128] first4: 0.646484 2.702148 -0.524902 1.139160
[Debug] dit_step18_vt_uncond: [689, 128] first4: 0.765625 2.664551 -0.493408 1.189453
[Debug] dit_step18_vt: [689, 128] first4: 0.563086 2.728467 -0.546948 1.103955
[Debug] dit_step18_xt: [689, 128] first4: 0.324790 2.435492 -0.489502 1.041656
[Debug] dit_step19_vt_cond: [689, 128] first4: 0.635254 2.832520 -0.547485 1.184570
[Debug] dit_step19_vt_uncond: [689, 128] first4: 0.751953 2.791992 -0.495361 1.235840
[Debug] dit_step19_vt: [689, 128] first4: 0.553564 2.860889 -0.583972 1.148682
[Debug] dit_step19_xt: [689, 128] first4: 0.343243 2.530855 -0.508968 1.079945
[Debug] dit_step20_vt_cond: [689, 128] first4: 0.606445 2.931641 -0.565752 1.227539
[Debug] dit_step20_vt_uncond: [689, 128] first4: 0.714844 2.875977 -0.499512 1.257324
[Debug] dit_step20_vt: [689, 128] first4: 0.530566 2.970605 -0.612120 1.206689
[Debug] dit_step20_xt: [689, 128] first4: 0.360928 2.629875 -0.529372 1.120168
[Debug] dit_step21_vt_cond: [689, 128] first4: 0.568848 3.013184 -0.578247 1.262207
[Debug] dit_step21_vt_uncond: [689, 128] first4: 0.672363 2.968018 -0.498352 1.284424
[Debug] dit_step21_vt: [689, 128] first4: 0.496387 3.044800 -0.634174 1.246655
[Debug] dit_step21_xt: [689, 128] first4: 0.377474 2.731369 -0.550511 1.161723
[Debug] dit_step22_vt_cond: [689, 128] first4: 0.535645 3.092773 -0.590088 1.305664
[Debug] dit_step22_vt_uncond: [689, 128] first4: 0.634277 3.045410 -0.503082 1.318848
[Debug] dit_step22_vt: [689, 128] first4: 0.466602 3.125928 -0.650992 1.296436
[Debug] dit_step22_xt: [689, 128] first4: 0.393028 2.835566 -0.572211 1.204938
[Debug] dit_step23_vt_cond: [689, 128] first4: 0.505371 3.171875 -0.599670 1.347900
[Debug] dit_step23_vt_uncond: [689, 128] first4: 0.595093 3.134033 -0.512939 1.351562
[Debug] dit_step23_vt: [689, 128] first4: 0.442566 3.198364 -0.660382 1.345337
[Debug] dit_step23_xt: [689, 128] first4: 0.407780 2.942178 -0.594223 1.249782
[Debug] dit_step24_vt_cond: [689, 128] first4: 0.486450 3.239990 -0.613159 1.397461
[Debug] dit_step24_vt_uncond: [689, 128] first4: 0.568115 3.175049 -0.531494 1.390320
[Debug] dit_step24_vt: [689, 128] first4: 0.429285 3.285449 -0.670325 1.402460
[Debug] dit_step24_xt: [689, 128] first4: 0.422089 3.051693 -0.616568 1.296531
[Debug] dit_step25_vt_cond: [689, 128] first4: 0.481995 3.310059 -0.628174 1.441895
[Debug] dit_step25_vt_uncond: [689, 128] first4: 0.545654 3.227539 -0.549805 1.416153
[Debug] dit_step25_vt: [689, 128] first4: 0.437433 3.367822 -0.683032 1.459914
[Debug] dit_step25_xt: [689, 128] first4: 0.436671 3.163954 -0.639335 1.345195
[Debug] dit_step26_vt_cond: [689, 128] first4: 0.490723 3.348267 -0.642090 1.473602
[Debug] dit_step26_vt_uncond: [689, 128] first4: 0.542114 3.329651 -0.576660 1.429016
[Debug] dit_step26_vt: [689, 128] first4: 0.454749 3.361298 -0.687891 1.504813
[Debug] dit_step26_xt: [689, 128] first4: 0.451829 3.275997 -0.662265 1.395355
[Debug] dit_step27_vt_cond: [689, 128] first4: 0.486572 3.388855 -0.674316 1.524292
[Debug] dit_step27_vt_uncond: [689, 128] first4: 0.534180 3.385681 -0.611816 1.466919
[Debug] dit_step27_vt: [689, 128] first4: 0.453247 3.391077 -0.718066 1.564453
[Debug] dit_step27_xt: [689, 128] first4: 0.466937 3.389033 -0.686201 1.447504
[Debug] dit_step28_vt_cond: [689, 128] first4: 0.500732 3.398682 -0.712158 1.541016
[Debug] dit_step28_vt_uncond: [689, 128] first4: 0.521729 3.410156 -0.660645 1.479248
[Debug] dit_step28_vt: [689, 128] first4: 0.486035 3.390649 -0.748218 1.584253
[Debug] dit_step28_xt: [689, 128] first4: 0.483138 3.502054 -0.711141 1.500312
[Debug] dit_step29_vt_cond: [689, 128] first4: 0.488281 3.448242 -0.766602 1.527954
[Debug] dit_step29_vt_uncond: [689, 128] first4: 0.511230 3.432373 -0.730469 1.473755
[Debug] dit_step29_vt: [689, 128] first4: 0.472217 3.459351 -0.791895 1.565894
[Debug] dit_step29_xt: [689, 128] first4: 0.498879 3.617366 -0.737538 1.552509
[Debug] dit_x0: [689, 128] first4: 0.498879 3.617366 -0.737538 1.552509
[Debug] window0_cond: [689, 2048] first4: 0.292676 -0.052296 0.071348 0.078284
[Debug] window0_latent: [689, 128] first4: 0.498879 3.617366 -0.737538 1.552509
[DiT] Window 1/1: T=689, 30 steps, 1908 ms (63.6 ms/step)
[DiT] CFG=1.70, 1 windows, 1.9 s
[VAE] Graph: 398 nodes, T_latent=689
[VAE] Decoded: T_latent=689 -> T_audio=352768 (8.00s @ 44.1kHz)
[Debug] vae_audio: [352768, 2] first4: -0.000973 -0.006419 -0.005822 -0.007101
[VAE] Decode: 1 windows -> 8.0s of audio, 111 ms
[Done] 34.7 s total
[Store] Unload VAE (103.4 MB)
[Store] Unload DiT (2477.0 MB)
[Store] Unload Cond (48.0 MB)
[WAV] Wrote ggml-dit/output.wav: 352768 samples, 44100 Hz, stereo
[Out] ggml-dit/output.wav
[Out] ggml-dit/output.json
Cannot initialize model with low cpu memory usage because `accelerate` was not found in the environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install `accelerate` for faster and less memory-intense model loading. You can do so with: 
```
pip install accelerate
```
.
Cannot initialize model with low cpu memory usage because `accelerate` was not found in the environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install `accelerate` for faster and less memory-intense model loading. You can do so with: 
```
pip install accelerate
```
.
[Request] Loaded request0.json
[DiT] steps=30, CFG=1.7 | mm3-dit-Q8_0.gguf
[GGML] Running MiniMax-Music3-transformer-Q8_0.gguf...
[GGML] Done, 135 dump files
[Python] Initializing transformer on cuda (T=689, 30 steps, CFG=1.7)...
[Python] Denoising...
[Python] Decoding audio...
[Python] Done, 104 dump files
[DiT] Cosine similarities GGML vs Python
noise: ggml vs python 1.000000
temb_t: ggml vs python 1.000000
hidden_after_preprocess: ggml vs python 1.000000
hidden_after_proj_in: ggml vs python 0.999993
layer0_sa_output: ggml vs python 0.999999
hidden_after_layer0: ggml vs python 0.999998
hidden_after_layer6: ggml vs python 1.000000
hidden_after_layer12: ggml vs python 1.000000
hidden_after_layer18: ggml vs python 0.999999
hidden_after_layer35: ggml vs python 0.999997
dit_step0_vt_cond: ggml vs python 0.999976
dit_step0_vt_uncond: ggml vs python 0.999961
dit_step0_vt: ggml vs python 0.999981
dit_step0_xt: ggml vs python 1.000000
dit_step5_vt_cond: ggml vs python 0.999961
dit_step5_vt: ggml vs python 0.999946
dit_step5_xt: ggml vs python 0.999995
dit_step10_vt_cond: ggml vs python 0.999952
dit_step10_vt: ggml vs python 0.999939
dit_step10_xt: ggml vs python 0.999983
dit_step15_vt_cond: ggml vs python 0.999941
dit_step15_vt: ggml vs python 0.999927
dit_step15_xt: ggml vs python 0.999974
dit_step20_vt_cond: ggml vs python 0.999920
dit_step20_vt: ggml vs python 0.999903
dit_step20_xt: ggml vs python 0.999967
dit_step25_vt_cond: ggml vs python 0.999914
dit_step25_vt: ggml vs python 0.999899
dit_step25_xt: ggml vs python 0.999961
dit_step29_vt_cond: ggml vs python 0.999923
dit_step29_vt: ggml vs python 0.999910
dit_x0: ggml vs python 0.999958
vae_audio: ggml vs python 0.999863
vae_audio (STFT cosine): ggml vs python 0.999919
[DiT] Error growth GGML vs Python
dit_step0_xt: cos 1.000000, max_err 0.002844, mean_err 0.000351, mean_A 0.000677, std_A 0.965967, mean_B 0.000641, std_B 0.965970
dit_step5_xt: cos 0.999995, max_err 0.022849, mean_err 0.002363, mean_A 0.012680, std_A 0.927385, mean_B 0.012502, std_B 0.927597
dit_step10_xt: cos 0.999983, max_err 0.067332, mean_err 0.004796, mean_A 0.024986, std_A 1.073370, mean_B 0.024634, std_B 1.073991
dit_step15_xt: cos 0.999974, max_err 0.096870, mean_err 0.007517, mean_A 0.037585, std_A 1.340587, mean_B 0.037044, std_B 1.341579
dit_step20_xt: cos 0.999967, max_err 0.139435, mean_err 0.010549, mean_A 0.050094, std_A 1.671618, mean_B 0.049350, std_B 1.672942
dit_step25_xt: cos 0.999961, max_err 0.189076, mean_err 0.013879, mean_A 0.062426, std_A 2.035708, mean_B 0.061458, std_B 2.037351