[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: CUDA0 (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: 1367 ms
[GGUF] ../models/MiniMax-Music3-rvq_depth_decoder-BF16.gguf: 47 tensors, data at offset 3456
[Load] Depth backend: CUDA0 (shared)
[WeightCtx] Loaded 47 tensors, 1232.4 MB into backend
[Depth] Loaded: 4 layers, dim 4096, 7 heads
[Store] Load Depth: 101 ms
[BPE] Loaded from GGUF: 151675 vocab, 151387 merges
[Prompt] 40 tokens
[AR] Prefill 70 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: -1.355042 -1.043213 -0.754630 0.104368
[AR] 200 frames total, 4.2 s (21.0 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: CUDA0 (CPU threads: 16)
[Cond] Loaded: mix over 8 states, proj 4096 -> 2048
[Store] Load Cond: 47 ms
[GGUF] ../models/MiniMax-Music3-transformer-Q4_K_M.gguf: 441 tensors, data at offset 31712
[Load] DiT backend: CUDA0 (shared)
[WeightCtx] Loaded 436 tensors, 1322.6 MB into backend
[DiT] Loaded: 36 layers, dim 2048, flash_attn=1
[Store] Load DiT: 112 ms
[GGUF] ../models/MiniMax-Music3-vocoder-F32.gguf: 182 tensors, data at offset 12896
[Load] VAE backend: CUDA0 (shared)
[VAE] Backend: CUDA0, Weight buffer: 103.4 MB
[VAE] Loaded: 4 blocks, upsample=512x, F32 activations
[Store] Load VAE: 103 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: 3.473374 -2.939100 -3.846277 3.896155
[Debug] hidden_after_preprocess: [689, 2304] first4: 0.341482 4.083567 -1.024049 1.790691
[Debug] hidden_after_proj_in: [689, 2048] first4: 5.319823 -4.328885 -6.038660 -1.736557
[Debug] layer0_sa_output: [690, 2048] first4: 1.310017 -1.718345 0.063078 1.546748
[Debug] hidden_after_layer0: [690, 2048] first4: -2.501665 -1.013151 -3.658370 3.644696
[Debug] hidden_after_layer6: [690, 2048] first4: -4.655267 -0.287687 0.026702 1.791035
[Debug] hidden_after_layer12: [690, 2048] first4: -2.504589 -1.578925 1.209268 3.278944
[Debug] hidden_after_layer18: [690, 2048] first4: -2.677608 -1.473281 -0.616087 1.134423
[Debug] hidden_after_layer35: [690, 2048] first4: -1.394914 0.351396 -4.714746 0.537621
[Debug] dit_step0_vt_cond: [689, 128] first4: -0.621199 -1.807535 -0.650863 -0.444658
[Debug] dit_step0_vt_uncond: [689, 128] first4: -0.274531 -2.211805 -0.471493 -0.666102
[Debug] dit_step0_vt: [689, 128] first4: -0.863867 -1.524546 -0.776422 -0.289647
[Debug] dit_step0_xt: [689, 128] first4: 0.165223 2.110556 -0.197931 0.839405
[Debug] dit_step1_vt_cond: [689, 128] first4: -0.747409 -1.780665 -0.590793 -0.381269
[Debug] dit_step1_vt_uncond: [689, 128] first4: -0.385127 -1.752235 -0.344022 -0.739594
[Debug] dit_step1_vt: [689, 128] first4: -1.001007 -1.800566 -0.763533 -0.130441
[Debug] dit_step1_xt: [689, 128] first4: 0.131856 2.050537 -0.223382 0.835057
[Debug] dit_step2_vt_cond: [689, 128] first4: -0.743548 -1.682285 -0.504244 -0.217875
[Debug] dit_step2_vt_uncond: [689, 128] first4: -0.710222 -1.953305 -0.623713 -0.720443
[Debug] dit_step2_vt: [689, 128] first4: -0.766877 -1.492571 -0.420616 0.133922
[Debug] dit_step2_xt: [689, 128] first4: 0.106294 2.000784 -0.237403 0.839521
[Debug] dit_step3_vt_cond: [689, 128] first4: -0.711629 -1.574798 -0.497576 -0.039270
[Debug] dit_step3_vt_uncond: [689, 128] first4: -0.359675 -1.834507 -0.645048 -0.398877
[Debug] dit_step3_vt: [689, 128] first4: -0.957996 -1.393001 -0.394346 0.212455
[Debug] dit_step3_xt: [689, 128] first4: 0.074361 1.954351 -0.250548 0.846603
[Debug] dit_step4_vt_cond: [689, 128] first4: -0.782477 -1.425823 -0.452518 0.190264
[Debug] dit_step4_vt_uncond: [689, 128] first4: -0.297932 -1.501732 -0.595393 -0.000975
[Debug] dit_step4_vt: [689, 128] first4: -1.121659 -1.372687 -0.352506 0.324131
[Debug] dit_step4_xt: [689, 128] first4: 0.036972 1.908595 -0.262298 0.857407
[Debug] dit_step5_vt_cond: [689, 128] first4: -0.827804 -1.285741 -0.428521 0.323969
[Debug] dit_step5_vt_uncond: [689, 128] first4: -0.442634 -1.291834 -0.551285 0.061322
[Debug] dit_step5_vt: [689, 128] first4: -1.097423 -1.281476 -0.342586 0.507822
[Debug] dit_step5_xt: [689, 128] first4: 0.000391 1.865879 -0.273718 0.874335
[Debug] dit_step6_vt_cond: [689, 128] first4: -0.902509 -1.155173 -0.408568 0.503355
[Debug] dit_step6_vt_uncond: [689, 128] first4: -0.450524 -1.214618 -0.660313 0.354511
[Debug] dit_step6_vt: [689, 128] first4: -1.218899 -1.113561 -0.232346 0.607545
[Debug] dit_step6_xt: [689, 128] first4: -0.040239 1.828760 -0.281462 0.894586
[Debug] dit_step7_vt_cond: [689, 128] first4: -0.953582 -0.965728 -0.307000 0.702887
[Debug] dit_step7_vt_uncond: [689, 128] first4: -0.630095 -1.083773 -0.753523 0.723318
[Debug] dit_step7_vt: [689, 128] first4: -1.180024 -0.883096 0.005566 0.688585
[Debug] dit_step7_xt: [689, 128] first4: -0.079573 1.799323 -0.281277 0.917539
[Debug] dit_step8_vt_cond: [689, 128] first4: -1.018095 -0.817609 -0.328189 0.818780
[Debug] dit_step8_vt_uncond: [689, 128] first4: -0.732260 -0.880573 -0.698589 0.950044
[Debug] dit_step8_vt: [689, 128] first4: -1.218178 -0.773534 -0.068910 0.726895
[Debug] dit_step8_xt: [689, 128] first4: -0.120179 1.773539 -0.283574 0.941769
[Debug] dit_step9_vt_cond: [689, 128] first4: -1.035028 -0.571669 -0.324888 0.947931
[Debug] dit_step9_vt_uncond: [689, 128] first4: -0.808856 -0.735822 -0.585598 1.074177
[Debug] dit_step9_vt: [689, 128] first4: -1.193348 -0.456763 -0.142392 0.859559
[Debug] dit_step9_xt: [689, 128] first4: -0.159957 1.758314 -0.288320 0.970421
[Debug] dit_step10_vt_cond: [689, 128] first4: -1.055950 -0.348806 -0.342800 1.082972
[Debug] dit_step10_vt_uncond: [689, 128] first4: -0.862926 -0.552315 -0.476209 1.154603
[Debug] dit_step10_vt: [689, 128] first4: -1.191067 -0.206350 -0.249414 1.032830
[Debug] dit_step10_xt: [689, 128] first4: -0.199659 1.751435 -0.296634 1.004849
[Debug] dit_step11_vt_cond: [689, 128] first4: -1.015361 -0.131280 -0.319553 1.201730
[Debug] dit_step11_vt_uncond: [689, 128] first4: -0.855245 -0.343387 -0.358908 1.259456
[Debug] dit_step11_vt: [689, 128] first4: -1.127443 0.017195 -0.292004 1.161322
[Debug] dit_step11_xt: [689, 128] first4: -0.237241 1.752008 -0.306367 1.043559
[Debug] dit_step12_vt_cond: [689, 128] first4: -1.009025 0.112310 -0.322910 1.319052
[Debug] dit_step12_vt_uncond: [689, 128] first4: -0.901670 -0.127652 -0.251058 1.311247
[Debug] dit_step12_vt: [689, 128] first4: -1.084174 0.280284 -0.373207 1.324515
[Debug] dit_step12_xt: [689, 128] first4: -0.273380 1.761351 -0.318808 1.087710
[Debug] dit_step13_vt_cond: [689, 128] first4: -1.092271 0.262591 -0.462590 1.364298
[Debug] dit_step13_vt_uncond: [689, 128] first4: -0.879953 0.154954 -0.187820 1.382699
[Debug] dit_step13_vt: [689, 128] first4: -1.240894 0.337936 -0.654928 1.351417
[Debug] dit_step13_xt: [689, 128] first4: -0.314743 1.772616 -0.340639 1.132757
[Debug] dit_step14_vt_cond: [689, 128] first4: -1.046313 0.498307 -0.461116 1.503875
[Debug] dit_step14_vt_uncond: [689, 128] first4: -0.920557 0.410305 -0.246307 1.476611
[Debug] dit_step14_vt: [689, 128] first4: -1.134343 0.559908 -0.611482 1.522960
[Debug] dit_step14_xt: [689, 128] first4: -0.352554 1.791279 -0.361021 1.183523
[Debug] dit_step15_vt_cond: [689, 128] first4: -1.035297 0.693956 -0.450204 1.623777
[Debug] dit_step15_vt_uncond: [689, 128] first4: -0.926898 0.646267 -0.220164 1.567225
[Debug] dit_step15_vt: [689, 128] first4: -1.111177 0.727338 -0.611232 1.663364
[Debug] dit_step15_xt: [689, 128] first4: -0.389594 1.815524 -0.381396 1.238968
[Debug] dit_step16_vt_cond: [689, 128] first4: -0.999638 0.904051 -0.412287 1.671841
[Debug] dit_step16_vt_uncond: [689, 128] first4: -0.941539 0.852787 -0.341528 1.601011
[Debug] dit_step16_vt: [689, 128] first4: -1.040306 0.939936 -0.461818 1.721421
[Debug] dit_step16_xt: [689, 128] first4: -0.424271 1.846855 -0.396790 1.296349
[Debug] dit_step17_vt_cond: [689, 128] first4: -0.994757 1.055006 -0.479309 1.717688
[Debug] dit_step17_vt_uncond: [689, 128] first4: -0.990737 1.077699 -0.323479 1.649263
[Debug] dit_step17_vt: [689, 128] first4: -0.997570 1.039121 -0.588389 1.765585
[Debug] dit_step17_xt: [689, 128] first4: -0.457523 1.881492 -0.416403 1.355202
[Debug] dit_step18_vt_cond: [689, 128] first4: -0.928169 1.284149 -0.396797 1.759125
[Debug] dit_step18_vt_uncond: [689, 128] first4: -0.927285 1.272468 -0.383827 1.627802
[Debug] dit_step18_vt: [689, 128] first4: -0.928788 1.292326 -0.405877 1.851052
[Debug] dit_step18_xt: [689, 128] first4: -0.488482 1.924570 -0.429932 1.416903
[Debug] dit_step19_vt_cond: [689, 128] first4: -0.867354 1.409108 -0.323001 1.802319
[Debug] dit_step19_vt_uncond: [689, 128] first4: -0.885613 1.502740 -0.240406 1.673116
[Debug] dit_step19_vt: [689, 128] first4: -0.854573 1.343566 -0.380817 1.892760
[Debug] dit_step19_xt: [689, 128] first4: -0.516968 1.969355 -0.442626 1.479995
[Debug] dit_step20_vt_cond: [689, 128] first4: -0.839404 1.561268 -0.445687 1.808472
[Debug] dit_step20_vt_uncond: [689, 128] first4: -0.736081 1.603020 -0.233248 1.707090
[Debug] dit_step20_vt: [689, 128] first4: -0.911731 1.532042 -0.594395 1.879439
[Debug] dit_step20_xt: [689, 128] first4: -0.547359 2.020424 -0.462439 1.542643
[Debug] dit_step21_vt_cond: [689, 128] first4: -0.731548 1.634565 -0.468722 1.758529
[Debug] dit_step21_vt_uncond: [689, 128] first4: -0.649751 1.729658 -0.255300 1.685258
[Debug] dit_step21_vt: [689, 128] first4: -0.788805 1.568000 -0.618118 1.809818
[Debug] dit_step21_xt: [689, 128] first4: -0.573653 2.072690 -0.483043 1.602971
[Debug] dit_step22_vt_cond: [689, 128] first4: -0.698780 1.811299 -0.463778 1.801114
[Debug] dit_step22_vt_uncond: [689, 128] first4: -0.651929 1.881573 -0.365998 1.699879
[Debug] dit_step22_vt: [689, 128] first4: -0.731576 1.762108 -0.532224 1.871978
[Debug] dit_step22_xt: [689, 128] first4: -0.598039 2.131427 -0.500784 1.665370
[Debug] dit_step23_vt_cond: [689, 128] first4: -0.671162 1.942481 -0.517307 1.763335
[Debug] dit_step23_vt_uncond: [689, 128] first4: -0.618647 1.971185 -0.381399 1.661531
[Debug] dit_step23_vt: [689, 128] first4: -0.707923 1.922388 -0.612443 1.834597
[Debug] dit_step23_xt: [689, 128] first4: -0.621636 2.195507 -0.521199 1.726523
[Debug] dit_step24_vt_cond: [689, 128] first4: -0.731492 2.038693 -0.554266 1.759919
[Debug] dit_step24_vt_uncond: [689, 128] first4: -0.596467 2.073166 -0.390361 1.682766
[Debug] dit_step24_vt: [689, 128] first4: -0.826009 2.014563 -0.668999 1.813926
[Debug] dit_step24_xt: [689, 128] first4: -0.649170 2.262659 -0.543499 1.786987
[Debug] dit_step25_vt_cond: [689, 128] first4: -0.668165 2.205301 -0.607180 1.780903
[Debug] dit_step25_vt_uncond: [689, 128] first4: -0.561425 2.183182 -0.432441 1.729295
[Debug] dit_step25_vt: [689, 128] first4: -0.742882 2.220785 -0.729498 1.817028
[Debug] dit_step25_xt: [689, 128] first4: -0.673932 2.336685 -0.567815 1.847555
[Debug] dit_step26_vt_cond: [689, 128] first4: -0.592792 2.350735 -0.569763 1.810312
[Debug] dit_step26_vt_uncond: [689, 128] first4: -0.555142 2.308111 -0.538856 1.707366
[Debug] dit_step26_vt: [689, 128] first4: -0.619146 2.380571 -0.591397 1.882374
[Debug] dit_step26_xt: [689, 128] first4: -0.694571 2.416038 -0.587528 1.910301
[Debug] dit_step27_vt_cond: [689, 128] first4: -0.551502 2.489485 -0.646134 1.823843
[Debug] dit_step27_vt_uncond: [689, 128] first4: -0.517987 2.426792 -0.564753 1.730777
[Debug] dit_step27_vt: [689, 128] first4: -0.574963 2.533370 -0.703101 1.888989
[Debug] dit_step27_xt: [689, 128] first4: -0.713736 2.500483 -0.610965 1.973267
[Debug] dit_step28_vt_cond: [689, 128] first4: -0.612787 2.573423 -0.725111 1.827590
[Debug] dit_step28_vt_uncond: [689, 128] first4: -0.586450 2.494137 -0.734498 1.731031
[Debug] dit_step28_vt: [689, 128] first4: -0.631224 2.628924 -0.718541 1.895181
[Debug] dit_step28_xt: [689, 128] first4: -0.734777 2.588114 -0.634916 2.036440
[Debug] dit_step29_vt_cond: [689, 128] first4: -0.630159 2.601351 -0.806841 1.746768
[Debug] dit_step29_vt_uncond: [689, 128] first4: -0.622233 2.576492 -0.809995 1.741788
[Debug] dit_step29_vt: [689, 128] first4: -0.635706 2.618752 -0.804634 1.750253
[Debug] dit_step29_xt: [689, 128] first4: -0.755967 2.675406 -0.661738 2.094781
[Debug] dit_x0: [689, 128] first4: -0.755967 2.675406 -0.661738 2.094781
[Debug] window0_cond: [689, 2048] first4: -0.290057 -0.003712 -0.019914 -0.070771
[Debug] window0_latent: [689, 128] first4: -0.755967 2.675406 -0.661738 2.094781
[DiT] Window 1/1: T=689, 30 steps, 2367 ms (78.9 ms/step)
[DiT] CFG=1.70, 1 windows, 2.4 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: 0.002068 -0.004974 -0.015524 0.001575
[VAE] Decode: 1 windows -> 8.0s of audio, 86 ms
[Done] 8.4 s total
[Store] Unload VAE (103.4 MB)
[Store] Unload DiT (1322.6 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-Q4_K_M.gguf
[GGML] Running MiniMax-Music3-transformer-Q4_K_M.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 0.999983
hidden_after_preprocess: ggml vs python 1.000000
hidden_after_proj_in: ggml vs python 0.999028
layer0_sa_output: ggml vs python 0.999788
hidden_after_layer0: ggml vs python 0.999828
hidden_after_layer6: ggml vs python 0.999909
hidden_after_layer12: ggml vs python 0.999858
hidden_after_layer18: ggml vs python 0.999841
hidden_after_layer35: ggml vs python 0.999470
dit_step0_vt_cond: ggml vs python 0.996623
dit_step0_vt_uncond: ggml vs python 0.996436
dit_step0_vt: ggml vs python 0.996399
dit_step0_xt: ggml vs python 0.999989
dit_step5_vt_cond: ggml vs python 0.992965
dit_step5_vt: ggml vs python 0.990210
dit_step5_xt: ggml vs python 0.999360
dit_step10_vt_cond: ggml vs python 0.991517
dit_step10_vt: ggml vs python 0.989683
dit_step10_xt: ggml vs python 0.997531
dit_step15_vt_cond: ggml vs python 0.990118
dit_step15_vt: ggml vs python 0.988631
dit_step15_xt: ggml vs python 0.995688
dit_step20_vt_cond: ggml vs python 0.988399
dit_step20_vt: ggml vs python 0.987134
dit_step20_xt: ggml vs python 0.994425
dit_step25_vt_cond: ggml vs python 0.987891
dit_step25_vt: ggml vs python 0.986981
dit_step25_xt: ggml vs python 0.993588
dit_step29_vt_cond: ggml vs python 0.988485
dit_step29_vt: ggml vs python 0.987778
dit_x0: ggml vs python 0.993164
vae_audio: ggml vs python 0.978284
vae_audio (STFT cosine): ggml vs python 0.988605
[DiT] Error growth GGML vs Python
dit_step0_xt: cos 0.999989, max_err 0.028576, mean_err 0.003582, mean_A -0.001110, std_A 0.965307, mean_B -0.002135, std_B 0.965041
dit_step5_xt: cos 0.999360, max_err 0.362212, mean_err 0.023928, mean_A 0.004516, std_A 0.884696, mean_B 0.002533, std_B 0.884593
dit_step10_xt: cos 0.997531, max_err 0.759452, mean_err 0.050026, mean_A 0.010681, std_A 0.946738, mean_B 0.007973, std_B 0.950289
dit_step15_xt: cos 0.995688, max_err 1.160728, mean_err 0.079050, mean_A 0.017418, std_A 1.124396, mean_B 0.013735, std_B 1.132820
dit_step20_xt: cos 0.994425, max_err 1.559546, mean_err 0.110387, mean_A 0.024243, std_A 1.372876, mean_B 0.019518, std_B 1.385859
dit_step25_xt: cos 0.993588, max_err 1.970941, mean_err 0.143886, mean_A 0.031048, std_A 1.660503, mean_B 0.025057, std_B 1.677655