civ
192 lines
15 KiB
Plaintext
192 lines
15 KiB
Plaintext
[Quant] BF16 -> ../models/omnivoice-base-BF16.gguf + ../models/omnivoice-tokenizer-BF16.gguf
|
|
[Input] Prompt: 102 chars: omnivoice.cpp is a minimal C++17 and GGML port of OmniVoice ...
|
|
[Input] RefText: 213 chars: If you go into different cultures, they have different conce...
|
|
[Input] RefWav: ../examples/freeman.wav
|
|
[Input] Language: English
|
|
[Input] Seed: 42
|
|
|
|
Loading weights: 0%| | 0/313 [00:00<?, ?it/s]
|
|
Loading weights: 100%|██████████| 313/313 [00:00<00:00, 29102.57it/s]
|
|
|
|
Loading weights: 0%| | 0/527 [00:00<?, ?it/s]
|
|
Loading weights: 100%|██████████| 527/527 [00:00<00:00, 30216.10it/s]
|
|
[CLI] Seed: 42
|
|
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/omnivoice.cpp/build/libggml-cuda.so
|
|
ggml_vulkan: Found 1 Vulkan devices:
|
|
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
|
|
load_backend: loaded Vulkan backend from /mnt/workspace/omnivoice.cpp/build/libggml-vulkan.so
|
|
load_backend: loaded CPU backend from /mnt/workspace/omnivoice.cpp/build/libggml-cpu-zen4.so
|
|
[Load] LM backend: Vulkan0 (CPU threads: 16)
|
|
[GGUF] ../models/omnivoice-base-BF16.gguf: 312 tensors, data at offset 5339136
|
|
[LM-Load] Loaded: 28L H=1024 FFN=3072 Nh=16 Nkv=8 D=128 theta=1000000 eps=1e-06 | audio: K=8 V=1025 mask_id=1024
|
|
[Qwen3] Attn: Q+K+V fused
|
|
[Qwen3] MLP: gate+up fused
|
|
[WeightCtx] Loaded 312 tensors, 1168.5 MB into backend
|
|
[Load] Flash attention disabled
|
|
[BPE] Loaded from GGUF: 151676 vocab, 151387 merges, eos_id=151643
|
|
[BPE] Registered 7 OmniVoice special tokens (total specials=8)
|
|
[GGUF] ../models/omnivoice-tokenizer-BF16.gguf: 486 tensors, data at offset 43680
|
|
[WeightCtx] Loaded 52 tensors, 11.1 MB into backend
|
|
[DAC] Loaded: 5 blocks, upsample 960x (8*5*4*2*3), 24 kHz mono out, weights 38.7 MB
|
|
[DAC-Enc] Loaded: 5 blocks, downsample 960x (8*5*4*2*3), 24 kHz mono in, weights 98.0 MB
|
|
[Sem-Enc] Loaded: 2 blocks, hidden=768, no temporal downsample, weights 28.1 MB
|
|
[HuBERT-Feat] Loaded: 7 conv layers, cumul stride 320, weights 8.0 MB
|
|
[HuBERT-Proj] Loaded: LN(512) + Linear(512 -> 768), weights 0.8 MB
|
|
[HuBERT-EncInit] Loaded: pos_conv k=128 g=16 + LN(768), weights 9.0 MB
|
|
[HuBERT-Stack] Loaded: 12 layers, hidden=768 heads=12 ffn=3072, weights 162.5 MB
|
|
[PipelineCodec] Loaded codec: sr=24000 hop=960 backend=Vulkan0
|
|
[WAV] Read buffer: 380672 samples, 22050 Hz, 1 ch, 16 bit
|
|
[Audio-Resample] 22050 Hz -> 24000 Hz, 380672 samples...
|
|
[Audio-Resample] Done: 380672 -> 414337 samples
|
|
[TTS] Reference: RMS 0.0271 -> 0.1 gain 3.6961
|
|
[TTS] Reference: silence-trim 414337 -> 414097 samples
|
|
[TTS] Reference: ../examples/freeman.wav, 414097 samples @ 24 kHz mono (17.25 s), aligned to 413760 (clip 337)
|
|
[Debug] ref-audio-16k: [275840] first4: -0.002174 -0.003920 -0.000047 -0.003989
|
|
[Debug] hubert-feat-extract: [512, 862] first4: 0.015111 0.002053 0.001422 0.000020
|
|
[Debug] hubert-feat-proj-ln: [862, 512] first4: 0.251028 -1.990119 -0.260102 -0.050774
|
|
[Debug] hubert-feat-proj: [862, 768] first4: -4.819709 -0.275990 0.393931 -0.054163
|
|
[Debug] hubert-enc-init: [862, 768] first4: 0.177062 -0.531735 0.651305 0.151246
|
|
[Debug] hubert-l0: [862, 768] first4: 0.392642 -0.978966 0.527780 -0.272625
|
|
[Debug] hubert-l5: [862, 768] first4: 0.135566 -0.454521 0.045241 -0.176600
|
|
[Debug] hubert-l7: [862, 768] first4: 0.128576 -0.472558 -0.166338 -0.342838
|
|
[Debug] hubert-l9: [862, 768] first4: 0.127733 -0.153881 0.266630 0.198758
|
|
[Debug] hubert-l11: [862, 768] first4: -0.071658 -0.068505 0.184706 0.373227
|
|
[Debug] ref-hubert-features: [431, 768] first4: 0.228764 -0.580392 0.248370 -0.091431
|
|
[TTS] Reference: encoded to [K=8, T=431] codes
|
|
[Debug] ref-audio-codes: [8, 431] first4: 694.000000 250.000000 283.000000 957.000000
|
|
[TTS-Long] Single-shot path: T=223 frames (8.92s), threshold=750 frames
|
|
[Prompt] Built: B'=2 K=8 S=738 N1=7 N2=77 Sref=431 Stgt=223 c_len=738 u_len=223 denoise=1
|
|
[Debug] prompt-cond-ids: [738] first4: 151669.000000 151670.000000 268.000000 151671.000000
|
|
[Debug] prompt-uncond-ids: [738] first4: 1024.000000 1024.000000 1024.000000 1024.000000
|
|
[TTS] Prompt: B'=2 K=8 S=738 c_len=738 u_len=223
|
|
[MaskGIT] Start: T=223 K=8 S=738 num_step=32 guidance=2.00 t_shift=0.100 layer_pen=5.00 cls_t=0.00 pos_t=5.00
|
|
[Debug] lm-hidden-step0-cond-embed: [738, 1024] first4: 0.075195 -0.010925 0.044434 0.095703
|
|
[Debug] lm-hidden-step0-cond-l0: [738, 1024] first4: -0.263336 0.469620 -0.249552 -2.017198
|
|
[Debug] lm-hidden-step0-cond-l1: [738, 1024] first4: 0.103641 0.098327 -0.419706 -4.095372
|
|
[Debug] lm-hidden-step0-cond-l2: [738, 1024] first4: 0.461552 -0.405512 -0.300792 -5.432191
|
|
[Debug] lm-hidden-step0-cond-l3: [738, 1024] first4: 0.314763 -0.284784 -0.300480 -5.575667
|
|
[Debug] lm-hidden-step0-cond-l4: [738, 1024] first4: 0.124147 -0.021900 -0.255000 -5.101854
|
|
[Debug] lm-hidden-step0-cond-l5: [738, 1024] first4: 0.671418 0.202003 -0.017993 -4.360538
|
|
[Debug] lm-hidden-step0-cond-l6: [738, 1024] first4: 0.439894 -0.508771 0.404440 -5.834112
|
|
[Debug] lm-hidden-step0-cond-l13: [738, 1024] first4: -3.047554 -3.972047 0.529790 -14.020215
|
|
[Debug] lm-hidden-step0-cond-l14: [738, 1024] first4: -3.717027 -4.369869 0.031428 -14.135784
|
|
[Debug] lm-hidden-step0-cond-l15: [738, 1024] first4: -3.897144 -4.523453 -0.456034 -15.353184
|
|
[Debug] lm-hidden-step0-cond-l16: [738, 1024] first4: -3.092649 -4.440215 -1.633492 -18.072735
|
|
[Debug] lm-hidden-step0-cond-l17: [738, 1024] first4: -4.218995 -4.688807 2.111674 -20.057915
|
|
[Debug] lm-hidden-step0-cond-l18: [738, 1024] first4: -2.167940 -4.824814 2.557554 -19.375731
|
|
[Debug] lm-hidden-step0-cond-l19: [738, 1024] first4: -3.482943 -3.029377 2.533296 -17.694117
|
|
[Debug] lm-hidden-step0-cond-l20: [738, 1024] first4: -3.523578 2.351006 8.961831 -15.824735
|
|
[Debug] lm-hidden-step0-cond-l1-norm1: [738, 1024] first4: -0.051857 0.133603 0.039453 0.157058
|
|
[Debug] lm-hidden-step0-cond-l1-attn: [738, 1024] first4: 0.368710 -0.012036 0.485296 -0.621393
|
|
[Debug] lm-hidden-step0-cond-l1-norm2: [738, 1024] first4: 0.080773 0.493726 -0.234968 -5.312475
|
|
[Debug] lm-hidden-step0-cond-l1-mlp: [738, 1024] first4: -0.001733 -0.359257 -0.655450 -1.456781
|
|
[Debug] lm-hidden-step0-cond: [738, 1024] first4: 0.152563 0.999222 16.983675 -0.232126
|
|
[Debug] lm-hidden-step0-uncond-embed: [738, 1024] first4: 0.005131 0.004723 0.009892 0.009244
|
|
[Debug] lm-hidden-step0-uncond-l0: [738, 1024] first4: -0.182689 -0.175698 0.261523 -2.662720
|
|
[Debug] lm-hidden-step0-uncond-l1: [738, 1024] first4: -0.930468 0.466411 0.382328 -2.862089
|
|
[Debug] lm-hidden-step0-uncond-l2: [738, 1024] first4: -1.487459 0.559479 0.573322 -3.368415
|
|
[Debug] lm-hidden-step0-uncond-l3: [738, 1024] first4: -0.798427 0.732602 0.488401 -3.312947
|
|
[Debug] lm-hidden-step0-uncond-l4: [738, 1024] first4: -0.436476 0.160034 0.331349 -3.361852
|
|
[Debug] lm-hidden-step0-uncond-l5: [738, 1024] first4: -0.152339 0.963246 -0.061818 -3.595951
|
|
[Debug] lm-hidden-step0-uncond-l6: [738, 1024] first4: -0.281944 1.535720 0.691676 -4.251910
|
|
[Debug] lm-hidden-step0-uncond-l13: [738, 1024] first4: 0.835587 1.044014 -0.216463 -11.851070
|
|
[Debug] lm-hidden-step0-uncond-l14: [738, 1024] first4: 0.473685 1.440260 0.051376 -12.754946
|
|
[Debug] lm-hidden-step0-uncond-l15: [738, 1024] first4: 1.368050 1.563073 0.901167 -15.908760
|
|
[Debug] lm-hidden-step0-uncond-l16: [738, 1024] first4: 1.598373 0.857502 0.078648 -15.258444
|
|
[Debug] lm-hidden-step0-uncond-l17: [738, 1024] first4: 1.343007 0.679198 1.510109 -14.588122
|
|
[Debug] lm-hidden-step0-uncond-l18: [738, 1024] first4: 1.083332 -0.768484 8.881135 -13.210123
|
|
[Debug] lm-hidden-step0-uncond-l19: [738, 1024] first4: 0.874424 2.609503 18.025383 -10.737823
|
|
[Debug] lm-hidden-step0-uncond-l20: [738, 1024] first4: 2.046572 9.887783 9.976087 -8.372459
|
|
[Debug] lm-hidden-step0-uncond-l1-norm1: [738, 1024] first4: -0.026022 -0.036154 -0.029905 0.149955
|
|
[Debug] lm-hidden-step0-uncond-l1-attn: [738, 1024] first4: -0.845916 -0.025419 0.574434 -0.198852
|
|
[Debug] lm-hidden-step0-uncond-l1-norm2: [738, 1024] first4: -0.539016 -0.148350 -0.569609 -3.938704
|
|
[Debug] lm-hidden-step0-uncond-l1-mlp: [738, 1024] first4: 0.098137 0.667528 -0.453629 -0.000517
|
|
[Debug] lm-hidden-step0-uncond: [738, 1024] first4: 1.354385 47.212521 129.784958 0.699243
|
|
[Debug] lm-logits-step0-cond: [8, 223, 1025] first4: 17.972193 13.713557 14.079697 15.216963
|
|
[Debug] lm-logits-step0-uncond: [8, 223, 1025] first4: 16.074760 12.336873 17.502558 15.306823
|
|
[Debug] mg-pred-tokens-step0: [8, 223] first4: 250.000000 739.000000 739.000000 739.000000
|
|
[Debug] mg-scores-step0: [8, 223] first4: -0.720804 -0.654032 -0.667639 -1.234109
|
|
[Debug] mg-log-probs-step0: [8, 223, 1025] first4: -22.480564 -27.780695 -37.013645 -29.210379
|
|
[MaskGIT-Step] 1/32 demask=6 remaining=1778
|
|
[MaskGIT-Step] 2/32 demask=7 remaining=1771
|
|
[MaskGIT-Step] 3/32 demask=7 remaining=1764
|
|
[MaskGIT-Step] 4/32 demask=7 remaining=1757
|
|
[MaskGIT-Step] 5/32 demask=8 remaining=1749
|
|
[MaskGIT-Step] 6/32 demask=8 remaining=1741
|
|
[MaskGIT-Step] 7/32 demask=9 remaining=1732
|
|
[MaskGIT-Step] 8/32 demask=9 remaining=1723
|
|
[MaskGIT-Step] 9/32 demask=10 remaining=1713
|
|
[MaskGIT-Step] 10/32 demask=11 remaining=1702
|
|
[MaskGIT-Step] 11/32 demask=12 remaining=1690
|
|
[MaskGIT-Step] 12/32 demask=13 remaining=1677
|
|
[MaskGIT-Step] 13/32 demask=14 remaining=1663
|
|
[MaskGIT-Step] 14/32 demask=15 remaining=1648
|
|
[MaskGIT-Step] 15/32 demask=16 remaining=1632
|
|
[MaskGIT-Step] 16/32 demask=18 remaining=1614
|
|
[MaskGIT-Step] 17/32 demask=20 remaining=1594
|
|
[MaskGIT-Step] 18/32 demask=22 remaining=1572
|
|
[MaskGIT-Step] 19/32 demask=25 remaining=1547
|
|
[MaskGIT-Step] 20/32 demask=28 remaining=1519
|
|
[MaskGIT-Step] 21/32 demask=32 remaining=1487
|
|
[MaskGIT-Step] 22/32 demask=36 remaining=1451
|
|
[MaskGIT-Step] 23/32 demask=42 remaining=1409
|
|
[MaskGIT-Step] 24/32 demask=49 remaining=1360
|
|
[MaskGIT-Step] 25/32 demask=58 remaining=1302
|
|
[MaskGIT-Step] 26/32 demask=70 remaining=1232
|
|
[MaskGIT-Step] 27/32 demask=87 remaining=1145
|
|
[MaskGIT-Step] 28/32 demask=110 remaining=1035
|
|
[MaskGIT-Step] 29/32 demask=143 remaining=892
|
|
[MaskGIT-Step] 30/32 demask=194 remaining=698
|
|
[MaskGIT-Step] 31/32 demask=279 remaining=419
|
|
[MaskGIT-Step] 32/32 demask=419 remaining=0
|
|
[MaskGIT] Total LM forward: 1520.96 ms across 32 steps (avg 47.53 ms/step)
|
|
[Debug] mg-tokens: [8, 223] first4: 997.000000 497.000000 205.000000 208.000000
|
|
[TTS] Decode: K=8 T=223 expected_samples=214080
|
|
[Debug] output-audio: [214080] first4: 0.001501 0.000549 0.002859 0.005375
|
|
[TTS-Long] Post-proc: 214080 -> 218880 samples (9.12s at 24000 Hz, ref_rms=0.0271)
|
|
[WAV] Wrote cpp/clone-cpp.wav: 218880 samples, 24000 Hz, mono F32
|
|
[OmniVoice-TTS] TTS: wrote cpp/clone-cpp.wav (218880 samples @ 24000 Hz, 9.12 s)
|
|
[Python] Audio: 218880 samples 9.12s -> python/clone-python.wav
|
|
[GGML] Cmd: ../build/omnivoice-tts --model ../models/omnivoice-base-BF16.gguf --codec ../models/omnivoice-tokenizer-BF16.gguf --seed 42 --ref-wav ../examples/freeman.wav --ref-text ../examples/freeman.txt --lang English --format wav32 --dump cpp -o cpp/clone-cpp.wav --no-fa
|
|
[GGML] Audio: 218880 samples 24000 Hz 9.12s -> cpp/clone-cpp.wav
|
|
[Cossim] PromptIDs cond exact: 99.59% diffs: 3 uncond exact: 100.00% diffs: 0
|
|
[Cossim] PromptIDs cond first diff range: s=84..193
|
|
[Cossim] RefAudio16k max_abs_diff: 3.323e-05 cossim: 1.000000 samples: 275840
|
|
[Cossim] HuBERT-features cossim: 0.999988 max_abs_diff: 1.626e-01 shape_cpp: (431, 768) shape_pt: (431, 768)
|
|
[Cossim] hubert-feat-extract cossim: 0.999978 max_abs_diff: 2.646e-03 shape_cpp: (512, 862) shape_pt: (512, 862)
|
|
[Cossim] hubert-feat-proj-ln cossim: 0.999986 max_abs_diff: 1.309e-01 shape_cpp: (862, 512) shape_pt: (862, 512)
|
|
[Cossim] hubert-feat-proj cossim: 0.999986 max_abs_diff: 1.023e-01 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] hubert-enc-init cossim: 0.999983 max_abs_diff: 2.472e-02 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] hubert-l0 cossim: 0.999984 max_abs_diff: 2.074e-02 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] hubert-l5 cossim: 0.999988 max_abs_diff: 2.133e-02 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] hubert-l7 cossim: 0.999989 max_abs_diff: 2.754e-02 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] hubert-l9 cossim: 0.999978 max_abs_diff: 1.916e-01 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] hubert-l11 cossim: 0.999939 max_abs_diff: 1.586e+00 shape_cpp: (862, 768) shape_pt: (862, 768)
|
|
[Cossim] RefAudioCodes exact: 95.45% diffs: 157 shape_cpp: (8, 431) shape_pt: (8, 431)
|
|
[Cossim] RefAudioCodes per-codebook: k0=99.3% k1=97.7% k2=96.5% k3=96.3% k4=95.1% k5=94.9% k6=91.6% k7=92.1%
|
|
[Cossim] L0 hidden cond: cos=0.993671 max_abs=7.657e+00 uncond: cos=0.999998 max_abs=3.941e-02
|
|
[Cossim] L1.norm1 hidden cond: cos=0.995029 max_abs=1.502e-01 uncond: cos=0.999997 max_abs=1.391e-03
|
|
[Cossim] L1.attn hidden cond: cos=0.999052 max_abs=1.519e-01 uncond: cos=0.999996 max_abs=6.630e-03
|
|
[Cossim] L1.norm2 hidden cond: cos=0.998567 max_abs=1.206e+00 uncond: cos=0.999998 max_abs=1.214e-02
|
|
[Cossim] L1.mlp hidden cond: cos=0.998562 max_abs=1.962e+00 uncond: cos=0.999994 max_abs=2.792e-02
|
|
[Cossim] L1 hidden cond: cos=0.995008 max_abs=8.024e+00 uncond: cos=0.999998 max_abs=4.992e-02
|
|
[Cossim] L2 hidden cond: cos=0.995600 max_abs=8.467e+00 uncond: cos=0.999998 max_abs=5.451e-02
|
|
[Cossim] L3 hidden cond: cos=0.995697 max_abs=8.898e+00 uncond: cos=0.999997 max_abs=5.960e-02
|
|
[Cossim] L4 hidden cond: cos=0.995768 max_abs=9.074e+00 uncond: cos=0.999996 max_abs=8.421e-02
|
|
[Cossim] L5 hidden cond: cos=0.995758 max_abs=9.719e+00 uncond: cos=0.999995 max_abs=8.283e-02
|
|
[Cossim] L6 hidden cond: cos=0.996028 max_abs=9.914e+00 uncond: cos=0.999994 max_abs=9.515e-02
|
|
[Cossim] L13 hidden cond: cos=0.997348 max_abs=1.508e+01 uncond: cos=0.999995 max_abs=1.805e-01
|
|
[Cossim] L14 hidden cond: cos=0.997905 max_abs=1.501e+01 uncond: cos=0.999994 max_abs=1.733e-01
|
|
[Cossim] L15 hidden cond: cos=0.998212 max_abs=1.511e+01 uncond: cos=0.999993 max_abs=3.689e-01
|
|
[Cossim] L16 hidden cond: cos=0.997945 max_abs=2.848e+01 uncond: cos=0.999993 max_abs=4.416e-01
|
|
[Cossim] L17 hidden cond: cos=0.998093 max_abs=2.851e+01 uncond: cos=0.999993 max_abs=4.348e-01
|
|
[Cossim] L18 hidden cond: cos=0.998404 max_abs=2.813e+01 uncond: cos=0.999993 max_abs=3.687e-01
|
|
[Cossim] L19 hidden cond: cos=0.999248 max_abs=2.953e+01 uncond: cos=0.999992 max_abs=5.512e-01
|
|
[Cossim] L20 hidden cond: cos=0.998969 max_abs=3.469e+01 uncond: cos=0.999992 max_abs=1.213e+00
|
|
[Cossim] Lf hidden cond: cos=0.999824 max_abs=1.267e+01 uncond: cos=0.999995 max_abs=8.033e-01
|
|
[Cossim] Logits cond: 1.000000 uncond: 1.000000
|
|
[Cossim] Step0 log_probs cossim: 0.999948 max_abs_diff: 3.369e+00
|
|
[Cossim] Step0 pred_tokens exact: 89.74% diffs: 183/1784
|
|
[Cossim] Step0 scores cossim: 0.999746 max_abs_diff: 3.213e-01
|
|
[Cossim] Tokens: 0.805904 exact: 19.96%
|
|
[Cossim] Audio: 0.277929
|
|
[Cossim] WAV stft_cos: 0.856300 samples: 218880
|