[Qwen] qwentts.cpp a62fde6 (2026-05-30)
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/qwentts.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/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-voicedesign-Q4_K_M.gguf: 404 tensors, data at offset 5949344
[WeightCtx] Loaded 316 tensors, 1012.7 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 109.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-Q4_K_M.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 16.4 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 25.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 13.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0 fa=off clamp_fp16=off
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.017703 -0.008537 0.002526 -0.014627
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] tts-pad-embed: [2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.364424 -0.020112 -0.018773 -0.087681
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -11.230798 0.703628 5.877054 -2.350159
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -10.802171 0.624090 6.020274 -2.465981
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -9.862975 -0.128177 6.852944 -1.702073
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -13.205417 -29.069054 8.021688 29.504890
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.618855 -1.422297 0.347773 1.260885
[Debug] talker-logits-prefill: [3072] first4: -5.714571 -9.181621 0.319536 -0.630142
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=2030 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=6 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=2026 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=359 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=421 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=810 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1124 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=288 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1502 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=706 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1484 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=85 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=558 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] next-emb-step0: [2048] first4: 0.036484 -0.058132 0.056991 0.061146
[Debug] talker-hidden-step1: [2048] first4: 1.593139 -1.844504 5.151402 -2.639456
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=85 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=875 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1018 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1430 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=278 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1751 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=405 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1440 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1870 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1358 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1502 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=783 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1956 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] output-audio: [122880] first4: 0.000081 0.000112 0.000114 0.000113
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] InstructIds shape: (1, 12)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] Q4_K_M -> ../models/qwen-talker-1.7b-voicedesign-Q4_K_M.gguf + ../models/qwen-tokenizer-12hz-Q4_K_M.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-Q4_K_M.gguf --codec ../models/qwen-tokenizer-12hz-Q4_K_M.gguf --seed 42 --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999666 max: 1.3890e-02 mean: 9.7826e-04
[Cossim] TrailingText cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] L0 cos: 0.999132 max: 1.0533e+00 mean: 6.7358e-03
[Cossim] L7 cos: 0.997443 max: 1.7684e+02 mean: 4.3788e-01
[Cossim] L14 cos: 0.997416 max: 1.7733e+02 mean: 5.4644e-01
[Cossim] L21 cos: 0.997017 max: 1.7519e+02 mean: 1.0163e+00
[Cossim] L27 cos: 0.985272 max: 6.2262e+02 mean: 4.7348e+00
[Cossim] Final cos: 0.932214 max: 2.7693e+01 mean: 5.1761e-01
[Cossim] Logits cos: 0.907321 max: 8.2766e+00 mean: 1.5713e+00
[Cossim] NextEmbStep0 cos: 0.803136 max: 1.8653e-01 mean: 3.3794e-02
[Cossim] TalkerHiddenStep1 cos: 0.741156 max: 9.0746e+00 mean: 1.5774e+00
[Cossim] CodesFull exact: 0.89% (1008 values)
[Cossim] Audio cos: 0.020622
[Cossim] WAV stft_cos: 0.227906 samples: 120960