[Request] Parsed ggml-turbo/request0.json [Request] seed=42 lm_batch=1 synth_batch=1 [Request] caption: An upbeat and anthemic pop-rock track driven by bright, slig... [Request] lyrics: 490 bytes [Request] bpm=83 dur=88 key=G major ts=4 lang=fr [Request] lm: temp=0.85 cfg=2.0 top_p=0.90 top_k=0 [Request] dit: steps=8 guidance=0.0 shift=3.0 [Request] task_type: text2music [Request] solver: euler (stork_substeps=10) [Request] lm_mode: generate [Request] output_format: mp3 [Request] synth_model: acestep-v15-turbo-Q5_K_M.gguf [Request] vae: vae-BF16.gguf [Request] audio_codes: (present) [Ace-Synth] Batch: 1 request(s) [Registry] Qwen3-Embedding-0.6B-BF16.gguf -> Text-Enc [Registry] Qwen3-Embedding-0.6B-Q8_0.gguf -> Text-Enc [Registry] acestep-5Hz-lm-0.6B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-0.6B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-4B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q5_K_M.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q6_K.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q8_0.gguf -> LM [Registry] acestep-v15-base-BF16.gguf -> DiT [Registry] acestep-v15-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-base-Q6_K.gguf -> DiT [Registry] acestep-v15-base-Q8_0.gguf -> DiT [Registry] acestep-v15-sft-BF16.gguf -> DiT [Registry] acestep-v15-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-BF16.gguf -> DiT [Registry] acestep-v15-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-continuous-BF16.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift1-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift3-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-base-BF16.gguf -> DiT [Registry] acestep-v15-xl-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-base-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sft-BF16.gguf -> DiT [Registry] acestep-v15-xl-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-turbo-BF16.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q8_0.gguf -> DiT [Registry] scragvae-BF16.gguf -> VAE [Registry] vae-BF16.gguf -> VAE [Store] Created (policy=STRICT) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-turbo-Q5_K_M.gguf: 678 tensors, data at offset 56864 [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-turbo-Q5_K_M.gguf: 678 tensors, data at offset 56864 [Synth-Load] Ready: turbo=yes, fa=no, batch_cfg=yes [Resolve-Params] max audio codes across batch: 434 (86.8s @ 5Hz) [Resolve-T] T=2170, S=1085 [Resolve-T] seed=42, steps=8, guidance=1.0, shift=3.0, duration=88.0s [BPE] Loaded from GGUF: 151643 vocab, 151387 merges 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/acestep.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/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] TextEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/Qwen3-Embedding-0.6B-BF16.gguf: 310 tensors, data at offset 5337568 [Load] TextEncoder: 28L, H=1024, Nh=16/8 [Qwen3] Attn: Q+K+V fused [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 310 tensors, 1136.5 MB into backend [Store] Load TextEnc: 5596 ms [Debug] text_hidden: [70, 1024] first4: 3.704526 2.436253 0.222853 -13.131872 [Debug] lyric_embed: [167, 1024] first4: 0.029175 0.032227 -0.022339 -0.028809 [Store] Unload TextEnc (1136.5 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] CondEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-turbo-Q5_K_M.gguf: 678 tensors, data at offset 56864 [Load] LyricEncoder: 8L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [Load] TimbreEncoder: 4L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 140 tensors, 412.5 MB into backend [Load] CondEncoder: lyric(8L), timbre(4L), text_proj, null_cond [Store] Load CondEnc: 676 ms [CondEnc] Lyric sliding mask: 167x167, window=128 [CondEnc] Timbre sliding mask: 1x1, window=128 [Encode] Packed: lyric=167 + timbre=1 + text=70 = 238 tokens [Encode-Text Batch0] 70+167 tokens -> enc_S=238, 121.1 ms [Debug] enc_hidden: [238, 2048] first4: 1.761670 -0.053200 -0.132888 0.058852 [Store] Unload CondEnc (412.5 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] Detokenizer backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-turbo-Q5_K_M.gguf: 678 tensors, data at offset 56864 [WeightCtx] Loaded 30 tensors, 73.2 MB into backend [Load] Detokenizer: FSQ(6->2048) + 2L encoder(S=5, 2048->64) [Store] Load FSQ-Detok: 60 ms [Context] Decoded: 434 codes -> 2170 frames (86.8s @ 25Hz) [Build-Context Batch0] Detokenizer: 413.4 ms, 434 codes [Debug] detok_output: [2170, 64] first4: -0.129311 1.458194 0.298132 -0.651512 [Store] Unload FSQ-Detok (73.2 MB) [Init-Noise Batch0] Philox noise seed=42, [2170, 64] solver=euler [Debug] noise: [2170, 64] first4: 0.194336 2.156250 -0.171875 0.847656 [Debug] context: [2170, 128] first4: -0.129311 1.458194 0.298132 -0.651512 [Init-Noise] Starting: T=2170, S=1085, enc_S=238, steps=8, batch=1 (cover) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] DiT backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-turbo-Q5_K_M.gguf: 678 tensors, data at offset 56864 [DiT] Self-attn: Q+K fused, V separate [DiT] Cross-attn: all separate [DiT] MLP: gate+up fused [Load] null_condition_emb found (CFG available) [WeightCtx] Loaded 478 tensors, 1061.2 MB into backend [Load] DiT: 24 layers, H=2048, Nh=16/8, D=128 [Store] Load DiT: 773 ms [DiT] Batch N=1, T=2170, S=1085, enc_S=238 [DiT] Graph: 2063 nodes [DiT] Solver: ODE Euler (1 NFE/step, order 1) [Debug] tproj: [12288] first4: 0.261152 -0.161305 -0.103153 0.050892 [Debug] temb: [2048] first4: -0.000119 -0.132132 -0.035650 0.065085 [Debug] temb_t: [2048] first4: 0.000588 0.026848 -0.052924 0.063878 [Debug] temb_r: [2048] first4: -0.000708 -0.158980 0.017274 0.001208 [Debug] sinusoidal_t: [256] first4: 0.562379 0.789627 0.439928 -0.023645 [Debug] sinusoidal_r: [256] first4: 1.000000 1.000000 1.000000 1.000000 [Debug] temb_lin1_t: [2048] first4: -0.051319 -0.053246 -0.011899 -0.038818 [Debug] temb_lin1_r: [2048] first4: -0.016165 -0.021121 -0.015801 -0.000525 [Debug] hidden_after_proj_in: [2048, 1085] first4: -1.048950 -0.942691 0.537616 0.450821 [Debug] proj_in_input: [192, 2170] first4: -0.129311 1.458194 0.298132 -0.651512 [Debug] enc_after_cond_emb: [2048, 238] first4: -0.145323 0.751763 0.348497 -0.533064 [Debug] layer0_sa_input: [2048, 1085] first4: -0.726623 -0.748099 -0.053174 0.262053 [Debug] layer0_q_after_rope: [128, 16] first4: -0.145323 0.751763 0.348497 -0.533064 [Debug] layer0_k_after_rope: [128, 8] first4: -1.551454 -1.000395 0.163224 0.462218 [Debug] layer0_sa_output: [2048, 1085] first4: -1.509493 0.130514 -0.385214 0.484345 [Debug] layer0_attn_out: [2048, 1085] first4: -1.551454 -1.000395 0.163224 0.462218 [Debug] layer0_after_self_attn: [2048, 1085] first4: -1.551454 -1.000395 0.163224 0.462218 [Debug] layer0_after_cross_attn: [2048, 1085] first4: -1.602293 -0.775263 -0.326485 0.501063 [Debug] hidden_after_layer0: [2048, 1085] first4: -9.293168 0.737673 51.618420 -0.869608 [Debug] hidden_after_layer6: [2048, 1085] first4: -20.846735 0.989748 29.713915 -4.707834 [Debug] hidden_after_layer12: [2048, 1085] first4: -18.078571 -17.109922 66.804359 24.826223 [Debug] hidden_after_layer18: [2048, 1085] first4: -26.631359 11.560244 58.145016 19.027023 [Debug] hidden_after_layer23: [2048, 1085] first4: -3.675964 36.309525 197.822388 157.329468 [Debug] dit_step0_vt: [2170, 64] first4: -0.002206 1.127751 0.286357 2.379969 [Debug] dit_step0_xt: [2170, 64] first4: 0.194436 2.104989 -0.184891 0.739476 [DiT] Step 1/8 t=1.000 [Debug] dit_step1_vt: [2170, 64] first4: -0.249564 1.364493 -0.159354 1.897966 [Debug] dit_step1_xt: [2170, 64] first4: 0.208049 2.030562 -0.176199 0.635950 [DiT] Step 2/8 t=0.955 [Debug] dit_step2_vt: [2170, 64] first4: -0.075789 1.247166 0.092347 2.368074 [Debug] dit_step2_xt: [2170, 64] first4: 0.213101 1.947417 -0.182356 0.478079 [DiT] Step 3/8 t=0.900 [Debug] dit_step3_vt: [2170, 64] first4: 0.249461 1.105958 0.234605 2.653776 [Debug] dit_step3_xt: [2170, 64] first4: 0.192313 1.855254 -0.201906 0.256931 [DiT] Step 4/8 t=0.833 [Debug] dit_step4_vt: [2170, 64] first4: 0.323155 0.867028 0.076578 2.721846 [Debug] dit_step4_xt: [2170, 64] first4: 0.157689 1.762358 -0.210111 -0.034695 [DiT] Step 5/8 t=0.750 [Debug] dit_step5_vt: [2170, 64] first4: 0.347711 0.761805 -0.411626 2.774529 [Debug] dit_step5_xt: [2170, 64] first4: 0.108016 1.653529 -0.151307 -0.431057 [DiT] Step 6/8 t=0.643 [Debug] dit_step6_vt: [2170, 64] first4: 0.182753 0.680159 -0.950754 2.783073 [Debug] dit_step6_xt: [2170, 64] first4: 0.071466 1.517497 0.038844 -0.987671 [DiT] Step 7/8 t=0.500 [Debug] dit_step7_vt: [2170, 64] first4: -0.012229 0.341252 -1.360142 3.065000 [Debug] dit_x0: [2170, 64] first4: 0.075134 1.415122 0.446886 -1.907171 [DiT] Step 8/8 t=0.300 [DiT-Generate] Total: 26984.3 ms (26984.3 ms/sample) [Debug] dit_output: [2170, 64] first4: 0.075134 1.415122 0.446886 -1.907171 [Store] Unload DiT (1061.2 MB) [GGUF] /mnt/workspace/git/acestep.cpp/models/vae-BF16.gguf: 365 tensors, data at offset 30048 load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] VAE backend: CPU (CPU threads: 16) [VAE] Backend: CPU, Weight buffer: 161.1 MB [VAE] Loaded: 5 blocks, upsample=1920x, F32 activations [Store] Load VAE-Dec: 728 ms [VAE] Tiled decode: 3 tiles (chunk=1024, overlap=64, stride=896) [VAE] Graph: 479 nodes, T_latent=960 [VAE] Upsample factor: 1920.00 (expected ~1920) [VAE] Graph: 479 nodes, T_latent=1024 [VAE] Graph: 479 nodes, T_latent=442 [VAE] Tiled decode done: 3 tiles -> T_audio=4166400 (86.80s @ 48kHz) [VAE-Decode Batch0] Decode: 32122.9 ms [Debug] vae_audio: [2, 4166400] first4: 0.000896 0.001622 0.001395 0.001934 [Store] Unload VAE-Dec (161.1 MB) [MP3] Encoding 86.8s @ 128 kbps, 48000 Hz stereo [MP3] 1388928 bytes (12.0:1), 496 ms (175.15x realtime), 32 threads [MP3] Wrote ggml-turbo/request00.mp3 [Ace-Synth] All done 2026-07-12 09:53:36.875 | DEBUG | acestep.core.generation.handler.init_service_memory_basic:_apply_malloc_mmap_threshold:50 - [memory] Set M_MMAP_THRESHOLD=131072 for immediate OS reclaim of large frees Skipping import of cpp extensions due to incompatible torch version. Please upgrade to torch >= 2.11.0 (found 2.10.0+cu128). 2026-07-12 09:53:39.012 | WARNING | acestep.training.trainer::40 - bitsandbytes not installed. Using standard AdamW. 2026-07-12 09:53:39.017 | INFO | acestep.gpu_config:get_gpu_memory_gb:578 - CUDA GPU detected: NVIDIA RTX PRO 6000 Blackwell Workstation Edition (95.0 GB) 2026-07-12 09:53:39.097 | INFO | acestep.core.generation.handler.init_service_loader:_load_main_model_from_checkpoint:174 - [initialize_service] Attempting to load model with attention implementation: sdpa Unable to import `torchao` Tensor objects. This may affect loading checkpoints serialized with `torchao` 2026-07-12 09:53:42.075 | INFO | acestep.core.generation.handler.generate_music:generate_music:289 - [generate_music] Turbo model detected: overriding guidance_scale 0.0 -> 1.0 (turbo does not use CFG). 2026-07-12 09:53:42.075 | INFO | acestep.core.generation.handler.generate_music:generate_music:304 - [generate_music] Starting generation... 2026-07-12 09:53:42.075 | INFO | acestep.core.generation.handler.generate_music:generate_music:307 - [generate_music] Preparing inputs... 2026-07-12 09:53:42.076 | INFO | acestep.core.generation.handler.generate_music_request:_prepare_reference_and_source_audio:179 - [generate_music] cover task: no src_audio but audio codes provided, proceeding with codes 2026-07-12 09:53:42.079 | INFO | acestep.core.generation.handler.generate_music:_vram_preflight_check:156 - [generate_music] VRAM pre-flight: 87.83 GB free, ~0.94 GB needed (batch=1, duration=88s, mode=turbo). 2026-07-12 09:53:42.085 | INFO | acestep.core.generation.handler.conditioning_target:_prepare_target_latents_and_wavs:78 - [generate_music] Decoding audio codes for item 0... 2026-07-12 09:53:42.290 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_precomputed_lm_hints:31 - [generate_music] Decoding audio codes for LM hints for item 0... 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:119 - ====================================================================== 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:120 - πŸ” [DEBUG] DiT TEXT ENCODER INPUT (Inference) 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:121 - ====================================================================== 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:122 - text_prompt: # Instruction Generate audio semantic tokens based on the given conditions: # Caption An upbeat and anthemic pop-rock track driven by bright, slightly overdriven # Metas - bpm: 83 - timesignature: 4 - keyscale: G major - duration: 88 seconds <|endoftext|> 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:123 - ====================================================================== 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:124 - lyrics_text: # Languages fr # Lyric # Lyric [Intro - Guitar Riff] [Verse 1] Dans le monde des tutos virtuels G ta toise en nouvelle passion Avec Ggendoline et PumbΓ© Γ  midi La communautΓ©, c'est l'unitΓ© Quel joie, une clΓ© [Chorus] Dans le monde des tutos virtuels GΓ’ndoline et PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner [Guitar Solo] [Verse 2] Dans le monde des tutos virtuels GΓ’ndoline, PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner<|endoftext|> 2026-07-12 09:53:42.293 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:125 - ====================================================================== 2026-07-12 09:53:42.300 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:110 - [preprocess_batch] Inferring prompt embeddings... 2026-07-12 09:53:42.315 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:113 - [preprocess_batch] Inferring lyric embeddings... 2026-07-12 09:53:42.315 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:161 - [service_generate] Generating audio... (DiT backend: PyTorch (cuda)) 2026-07-12 09:53:42.338 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:263 - [service_generate] DiT diffusion via PyTorch (cuda)... 2026-07-12 09:53:42.346 | INFO | acestep.models.common.dcw_correction:__init__:98 - [DCW] Active β€” mode=double, scaler=0.0500, high_scaler=0.0200, wavelet='haar' 2026-07-12 09:53:42.447 | WARNING | acestep.models.common.dcw_loader:_try_import:39 - DCW is enabled but 'pytorch_wavelets' is not installed. Install with `pip install pytorch_wavelets PyWavelets` to use Differential Correction in Wavelet domain. Falling back to no-op for this generation. 2026-07-12 09:53:42.671 | INFO | acestep.core.generation.handler.generate_music_decode:_prepare_generate_music_decode_state:42 - [generate_music] Model generation completed. Decoding latents... 2026-07-12 09:53:42.680 | DEBUG | acestep.core.generation.handler.generate_music_decode:_prepare_generate_music_decode_state:64 - [generate_music] pred_latents: torch.Size([1, 2170, 64]), dtype=torch.bfloat16 2026-07-12 09:53:42.680 | DEBUG | acestep.core.generation.handler.generate_music_decode:_prepare_generate_music_decode_state:65 - [generate_music] time_costs: {'encoder_time_cost': 0.006998777389526367, 'diffusion_time_cost': 0.3245120048522949, 'diffusion_per_step_time_cost': 0.040564000606536865, 'total_time_cost': 0.3315107822418213, 'offload_time_cost': 0.0} 2026-07-12 09:53:42.685 | INFO | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:127 - [generate_music] Decoding latents with VAE... 2026-07-12 09:53:42.688 | DEBUG | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:136 - [generate_music] Before VAE decode: allocated=5.98GB, max=7.29GB 2026-07-12 09:53:42.688 | INFO | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:154 - [generate_music] Effective free VRAM before VAE decode: 87.72 GB 2026-07-12 09:53:42.688 | INFO | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:172 - [generate_music] Using tiled VAE decode to reduce VRAM usage... 2026-07-12 09:53:42.688 | DEBUG | acestep.core.generation.handler.memory_utils:_get_auto_decode_chunk_size:108 - [_get_auto_decode_chunk_size] Effective free VRAM: 87.72 GB 2026-07-12 09:53:42.688 | DEBUG | acestep.core.generation.handler.memory_utils:_should_offload_wav_to_cpu:131 - [_should_offload_wav_to_cpu] Effective free VRAM: 87.72 GB 2026-07-12 09:53:42.688 | INFO | acestep.core.generation.handler.vae_decode:tiled_decode:56 - [tiled_decode] chunk_size=512, offload_wav_to_cpu=False, latents_shape=torch.Size([1, 64, 2170]) 2026-07-12 09:53:42.988 | DEBUG | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:193 - [generate_music] After VAE decode: allocated=6.15GB, max=7.44GB 2026-07-12 09:53:43.155 | INFO | acestep.core.generation.handler.generate_music_payload:_build_generate_music_success_payload:38 - [generate_music] VAE decode completed. Preparing audio tensors... 2026-07-12 09:53:43.159 | INFO | acestep.core.generation.handler.generate_music_payload:_build_generate_music_success_payload:50 - [generate_music] Done! Generated 1 audio tensors. [Request] Parsed ggml-sft/request0.json [Request] seed=42 lm_batch=1 synth_batch=1 [Request] caption: An upbeat and anthemic pop-rock track driven by bright, slig... [Request] lyrics: 490 bytes [Request] bpm=83 dur=88 key=G major ts=4 lang=fr [Request] lm: temp=0.85 cfg=2.0 top_p=0.90 top_k=0 [Request] dit: steps=50 guidance=1.0 shift=1.0 [Request] task_type: text2music [Request] solver: euler (stork_substeps=10) [Request] lm_mode: generate [Request] output_format: mp3 [Request] synth_model: acestep-v15-sft-Q5_K_M.gguf [Request] vae: vae-BF16.gguf [Request] audio_codes: (present) [Ace-Synth] Batch: 1 request(s) [Registry] Qwen3-Embedding-0.6B-BF16.gguf -> Text-Enc [Registry] Qwen3-Embedding-0.6B-Q8_0.gguf -> Text-Enc [Registry] acestep-5Hz-lm-0.6B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-0.6B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-4B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q5_K_M.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q6_K.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q8_0.gguf -> LM [Registry] acestep-v15-base-BF16.gguf -> DiT [Registry] acestep-v15-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-base-Q6_K.gguf -> DiT [Registry] acestep-v15-base-Q8_0.gguf -> DiT [Registry] acestep-v15-sft-BF16.gguf -> DiT [Registry] acestep-v15-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-BF16.gguf -> DiT [Registry] acestep-v15-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-continuous-BF16.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift1-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift3-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-base-BF16.gguf -> DiT [Registry] acestep-v15-xl-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-base-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sft-BF16.gguf -> DiT [Registry] acestep-v15-xl-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-turbo-BF16.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q8_0.gguf -> DiT [Registry] scragvae-BF16.gguf -> VAE [Registry] vae-BF16.gguf -> VAE [Store] Created (policy=STRICT) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-sft-Q5_K_M.gguf: 678 tensors, data at offset 56800 [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-sft-Q5_K_M.gguf: 678 tensors, data at offset 56800 [Synth-Load] Ready: turbo=no, fa=no, batch_cfg=yes [Resolve-Params] max audio codes across batch: 434 (86.8s @ 5Hz) [Resolve-T] T=2170, S=1085 [Resolve-T] seed=42, steps=50, guidance=1.0, shift=1.0, duration=88.0s [BPE] Loaded from GGUF: 151643 vocab, 151387 merges 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/acestep.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/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] TextEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/Qwen3-Embedding-0.6B-BF16.gguf: 310 tensors, data at offset 5337568 [Load] TextEncoder: 28L, H=1024, Nh=16/8 [Qwen3] Attn: Q+K+V fused [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 310 tensors, 1136.5 MB into backend [Store] Load TextEnc: 255 ms [Debug] text_hidden: [70, 1024] first4: 3.704526 2.436253 0.222853 -13.131872 [Debug] lyric_embed: [167, 1024] first4: 0.029175 0.032227 -0.022339 -0.028809 [Store] Unload TextEnc (1136.5 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] CondEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-sft-Q5_K_M.gguf: 678 tensors, data at offset 56800 [Load] LyricEncoder: 8L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [Load] TimbreEncoder: 4L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 140 tensors, 412.5 MB into backend [Load] CondEncoder: lyric(8L), timbre(4L), text_proj, null_cond [Store] Load CondEnc: 688 ms [CondEnc] Lyric sliding mask: 167x167, window=128 [CondEnc] Timbre sliding mask: 1x1, window=128 [Encode] Packed: lyric=167 + timbre=1 + text=70 = 238 tokens [Encode-Text Batch0] 70+167 tokens -> enc_S=238, 122.9 ms [Debug] enc_hidden: [238, 2048] first4: 1.761670 -0.053200 -0.132888 0.058852 [Store] Unload CondEnc (412.5 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] Detokenizer backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-sft-Q5_K_M.gguf: 678 tensors, data at offset 56800 [WeightCtx] Loaded 30 tensors, 73.2 MB into backend [Load] Detokenizer: FSQ(6->2048) + 2L encoder(S=5, 2048->64) [Store] Load FSQ-Detok: 60 ms [Context] Decoded: 434 codes -> 2170 frames (86.8s @ 25Hz) [Build-Context Batch0] Detokenizer: 416.5 ms, 434 codes [Debug] detok_output: [2170, 64] first4: -0.129311 1.458194 0.298132 -0.651512 [Store] Unload FSQ-Detok (73.2 MB) [Init-Noise Batch0] Philox noise seed=42, [2170, 64] solver=euler [Debug] noise: [2170, 64] first4: 0.194336 2.156250 -0.171875 0.847656 [Debug] context: [2170, 128] first4: -0.129311 1.458194 0.298132 -0.651512 [Init-Noise] Starting: T=2170, S=1085, enc_S=238, steps=50, batch=1 (cover) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] DiT backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-sft-Q5_K_M.gguf: 678 tensors, data at offset 56800 [DiT] Self-attn: Q+K fused, V separate [DiT] Cross-attn: all separate [DiT] MLP: gate+up fused [Load] null_condition_emb found (CFG available) [WeightCtx] Loaded 478 tensors, 1061.2 MB into backend [Load] DiT: 24 layers, H=2048, Nh=16/8, D=128 [Store] Load DiT: 786 ms [DiT] Batch N=1, T=2170, S=1085, enc_S=238 [DiT] Graph: 2063 nodes [DiT] Solver: ODE Euler (1 NFE/step, order 1) [Debug] tproj: [12288] first4: 0.155936 -0.117620 -0.084492 0.081893 [Debug] temb: [2048] first4: -0.002754 -0.176678 0.004687 -0.001972 [Debug] temb_t: [2048] first4: -0.001256 0.003101 -0.012669 -0.001763 [Debug] temb_r: [2048] first4: -0.001498 -0.179779 0.017356 -0.000209 [Debug] sinusoidal_t: [256] first4: 0.562379 0.789627 0.439928 -0.023645 [Debug] sinusoidal_r: [256] first4: 1.000000 1.000000 1.000000 1.000000 [Debug] temb_lin1_t: [2048] first4: -0.034934 0.029198 0.027409 -0.031502 [Debug] temb_lin1_r: [2048] first4: -0.002680 0.004202 0.000655 -0.002088 [Debug] hidden_after_proj_in: [2048, 1085] first4: -1.099326 -0.899651 0.499825 0.444552 [Debug] proj_in_input: [192, 2170] first4: -0.129311 1.458194 0.298132 -0.651512 [Debug] enc_after_cond_emb: [2048, 238] first4: -0.166779 0.841212 0.381438 -0.573819 [Debug] layer0_sa_input: [2048, 1085] first4: -0.922690 -0.700908 -0.034981 0.295921 [Debug] layer0_q_after_rope: [128, 16] first4: -0.166779 0.841212 0.381438 -0.573819 [Debug] layer0_k_after_rope: [128, 8] first4: -1.796069 -1.336702 -0.152231 0.385512 [Debug] layer0_sa_output: [2048, 1085] first4: -1.616210 0.770564 -0.580455 0.537726 [Debug] layer0_attn_out: [2048, 1085] first4: -1.796069 -1.336702 -0.152231 0.385512 [Debug] layer0_after_self_attn: [2048, 1085] first4: -1.796069 -1.336702 -0.152231 0.385512 [Debug] layer0_after_cross_attn: [2048, 1085] first4: -1.972029 -1.019393 -0.413236 0.394381 [Debug] hidden_after_layer0: [2048, 1085] first4: -9.690948 1.413015 60.449055 -1.652728 [Debug] hidden_after_layer6: [2048, 1085] first4: -16.032001 2.777304 60.309673 -1.417665 [Debug] hidden_after_layer12: [2048, 1085] first4: -13.579681 2.601818 -15.388254 -0.629633 [Debug] hidden_after_layer18: [2048, 1085] first4: -6.477047 16.179531 -40.314045 -11.229168 [Debug] hidden_after_layer23: [2048, 1085] first4: 35.650501 64.484314 52.944641 -18.251856 [Debug] dit_step0_vt: [2170, 64] first4: -0.657627 2.496069 -0.170021 1.409376 [Debug] dit_step0_xt: [2170, 64] first4: 0.207488 2.106329 -0.168475 0.819469 [DiT] Step 1/50 t=1.000 [Debug] dit_step1_vt: [2170, 64] first4: -0.714396 2.509334 -0.058743 1.236221 [Debug] dit_step1_xt: [2170, 64] first4: 0.221776 2.056142 -0.167300 0.794744 [DiT] Step 2/50 t=0.980 [Debug] dit_step2_vt: [2170, 64] first4: -0.697460 2.508775 0.034986 1.306383 [Debug] dit_step2_xt: [2170, 64] first4: 0.235726 2.005966 -0.167999 0.768617 [DiT] Step 3/50 t=0.960 [Debug] dit_step3_vt: [2170, 64] first4: -0.697118 2.415090 0.037855 1.297944 [Debug] dit_step3_xt: [2170, 64] first4: 0.249668 1.957665 -0.168757 0.742658 [DiT] Step 4/50 t=0.940 [Debug] dit_step4_vt: [2170, 64] first4: -0.602095 2.320011 -0.002184 1.288608 [Debug] dit_step4_xt: [2170, 64] first4: 0.261710 1.911264 -0.168713 0.716886 [DiT] Step 5/50 t=0.920 [Debug] dit_step5_vt: [2170, 64] first4: -0.589762 2.295776 0.016944 1.278431 [Debug] dit_step5_xt: [2170, 64] first4: 0.273505 1.865349 -0.169052 0.691317 [DiT] Step 6/50 t=0.900 [Debug] dit_step6_vt: [2170, 64] first4: -0.574608 2.234402 0.000571 1.288834 [Debug] dit_step6_xt: [2170, 64] first4: 0.284997 1.820661 -0.169063 0.665540 [DiT] Step 7/50 t=0.880 [Debug] dit_step7_vt: [2170, 64] first4: -0.533204 2.178506 -0.056891 1.280819 [Debug] dit_step7_xt: [2170, 64] first4: 0.295661 1.777091 -0.167925 0.639924 [DiT] Step 8/50 t=0.860 [Debug] dit_step8_vt: [2170, 64] first4: -0.455220 2.094212 -0.047918 1.294091 [Debug] dit_step8_xt: [2170, 64] first4: 0.304766 1.735207 -0.166967 0.614042 [DiT] Step 9/50 t=0.840 [Debug] dit_step9_vt: [2170, 64] first4: -0.381123 1.974700 -0.038506 1.269894 [Debug] dit_step9_xt: [2170, 64] first4: 0.312388 1.695713 -0.166197 0.588644 [DiT] Step 10/50 t=0.820 [Debug] dit_step10_vt: [2170, 64] first4: -0.354808 1.865164 -0.019080 1.252590 [Debug] dit_step10_xt: [2170, 64] first4: 0.319484 1.658409 -0.165815 0.563592 [DiT] Step 11/50 t=0.800 [Debug] dit_step11_vt: [2170, 64] first4: -0.327874 1.788247 -0.006149 1.241280 [Debug] dit_step11_xt: [2170, 64] first4: 0.326042 1.622644 -0.165692 0.538767 [DiT] Step 12/50 t=0.780 [Debug] dit_step12_vt: [2170, 64] first4: -0.277828 1.760163 0.018898 1.222847 [Debug] dit_step12_xt: [2170, 64] first4: 0.331598 1.587441 -0.166070 0.514310 [DiT] Step 13/50 t=0.760 [Debug] dit_step13_vt: [2170, 64] first4: -0.267371 1.683511 -0.038392 1.187180 [Debug] dit_step13_xt: [2170, 64] first4: 0.336946 1.553771 -0.165302 0.490566 [DiT] Step 14/50 t=0.740 [Debug] dit_step14_vt: [2170, 64] first4: -0.249646 1.585912 -0.031737 1.160405 [Debug] dit_step14_xt: [2170, 64] first4: 0.341939 1.522053 -0.164668 0.467358 [DiT] Step 15/50 t=0.720 [Debug] dit_step15_vt: [2170, 64] first4: -0.273806 1.517243 -0.054140 1.149361 [Debug] dit_step15_xt: [2170, 64] first4: 0.347415 1.491708 -0.163585 0.444371 [DiT] Step 16/50 t=0.700 [Debug] dit_step16_vt: [2170, 64] first4: -0.276138 1.435699 -0.078307 1.123911 [Debug] dit_step16_xt: [2170, 64] first4: 0.352938 1.462994 -0.162019 0.421893 [DiT] Step 17/50 t=0.680 [Debug] dit_step17_vt: [2170, 64] first4: -0.256064 1.342587 -0.043681 1.153011 [Debug] dit_step17_xt: [2170, 64] first4: 0.358059 1.436142 -0.161145 0.398833 [DiT] Step 18/50 t=0.660 [Debug] dit_step18_vt: [2170, 64] first4: -0.267567 1.261434 -0.113456 1.137467 [Debug] dit_step18_xt: [2170, 64] first4: 0.363410 1.410913 -0.158876 0.376083 [DiT] Step 19/50 t=0.640 [Debug] dit_step19_vt: [2170, 64] first4: -0.284692 1.182765 -0.130472 1.109357 [Debug] dit_step19_xt: [2170, 64] first4: 0.369104 1.387258 -0.156267 0.353896 [DiT] Step 20/50 t=0.620 [Debug] dit_step20_vt: [2170, 64] first4: -0.337225 1.075767 -0.134062 1.043689 [Debug] dit_step20_xt: [2170, 64] first4: 0.375849 1.365743 -0.153585 0.333022 [DiT] Step 21/50 t=0.600 [Debug] dit_step21_vt: [2170, 64] first4: -0.361715 1.045356 -0.102544 1.084714 [Debug] dit_step21_xt: [2170, 64] first4: 0.383083 1.344836 -0.151534 0.311328 [DiT] Step 22/50 t=0.580 [Debug] dit_step22_vt: [2170, 64] first4: -0.357367 0.932388 -0.160807 1.090126 [Debug] dit_step22_xt: [2170, 64] first4: 0.390230 1.326188 -0.148318 0.289526 [DiT] Step 23/50 t=0.560 [Debug] dit_step23_vt: [2170, 64] first4: -0.348208 0.853325 -0.160243 1.061289 [Debug] dit_step23_xt: [2170, 64] first4: 0.397194 1.309121 -0.145113 0.268300 [DiT] Step 24/50 t=0.540 [Debug] dit_step24_vt: [2170, 64] first4: -0.406379 0.755160 -0.132647 0.963357 [Debug] dit_step24_xt: [2170, 64] first4: 0.405322 1.294018 -0.142460 0.249033 [DiT] Step 25/50 t=0.520 [Debug] dit_step25_vt: [2170, 64] first4: -0.445441 0.659720 -0.139801 0.969728 [Debug] dit_step25_xt: [2170, 64] first4: 0.414231 1.280824 -0.139664 0.229638 [DiT] Step 26/50 t=0.500 [Debug] dit_step26_vt: [2170, 64] first4: -0.472626 0.550692 -0.093925 0.991653 [Debug] dit_step26_xt: [2170, 64] first4: 0.423683 1.269810 -0.137786 0.209805 [DiT] Step 27/50 t=0.480 [Debug] dit_step27_vt: [2170, 64] first4: -0.480618 0.430827 -0.139776 0.935210 [Debug] dit_step27_xt: [2170, 64] first4: 0.433296 1.261193 -0.134990 0.191101 [DiT] Step 28/50 t=0.460 [Debug] dit_step28_vt: [2170, 64] first4: -0.534665 0.298242 -0.164471 0.880820 [Debug] dit_step28_xt: [2170, 64] first4: 0.443989 1.255228 -0.131701 0.173484 [DiT] Step 29/50 t=0.440 [Debug] dit_step29_vt: [2170, 64] first4: -0.592525 0.213271 -0.131122 0.905263 [Debug] dit_step29_xt: [2170, 64] first4: 0.455839 1.250963 -0.129079 0.155379 [DiT] Step 30/50 t=0.420 [Debug] dit_step30_vt: [2170, 64] first4: -0.590435 0.124735 -0.086982 0.866790 [Debug] dit_step30_xt: [2170, 64] first4: 0.467648 1.248468 -0.127339 0.138043 [DiT] Step 31/50 t=0.400 [Debug] dit_step31_vt: [2170, 64] first4: -0.561385 0.003844 -0.094693 0.840834 [Debug] dit_step31_xt: [2170, 64] first4: 0.478876 1.248391 -0.125445 0.121227 [DiT] Step 32/50 t=0.380 [Debug] dit_step32_vt: [2170, 64] first4: -0.603585 -0.073918 -0.072624 0.812255 [Debug] dit_step32_xt: [2170, 64] first4: 0.490948 1.249870 -0.123993 0.104982 [DiT] Step 33/50 t=0.360 [Debug] dit_step33_vt: [2170, 64] first4: -0.566747 -0.235732 -0.025024 0.803259 [Debug] dit_step33_xt: [2170, 64] first4: 0.502283 1.254584 -0.123492 0.088916 [DiT] Step 34/50 t=0.340 [Debug] dit_step34_vt: [2170, 64] first4: -0.591787 -0.320321 -0.085411 0.788053 [Debug] dit_step34_xt: [2170, 64] first4: 0.514118 1.260991 -0.121784 0.073155 [DiT] Step 35/50 t=0.320 [Debug] dit_step35_vt: [2170, 64] first4: -0.629666 -0.432637 -0.049729 0.764802 [Debug] dit_step35_xt: [2170, 64] first4: 0.526712 1.269644 -0.120789 0.057859 [DiT] Step 36/50 t=0.300 [Debug] dit_step36_vt: [2170, 64] first4: -0.586688 -0.514378 0.023927 0.698038 [Debug] dit_step36_xt: [2170, 64] first4: 0.538445 1.279931 -0.121268 0.043899 [DiT] Step 37/50 t=0.280 [Debug] dit_step37_vt: [2170, 64] first4: -0.600946 -0.582601 0.050672 0.700169 [Debug] dit_step37_xt: [2170, 64] first4: 0.550464 1.291583 -0.122281 0.029895 [DiT] Step 38/50 t=0.260 [Debug] dit_step38_vt: [2170, 64] first4: -0.641691 -0.674089 0.034943 0.665093 [Debug] dit_step38_xt: [2170, 64] first4: 0.563298 1.305065 -0.122980 0.016593 [DiT] Step 39/50 t=0.240 [Debug] dit_step39_vt: [2170, 64] first4: -0.577388 -0.770592 0.072429 0.594378 [Debug] dit_step39_xt: [2170, 64] first4: 0.574846 1.320477 -0.124429 0.004706 [DiT] Step 40/50 t=0.220 [Debug] dit_step40_vt: [2170, 64] first4: -0.574397 -0.861981 0.093438 0.622612 [Debug] dit_step40_xt: [2170, 64] first4: 0.586334 1.337716 -0.126297 -0.007746 [DiT] Step 41/50 t=0.200 [Debug] dit_step41_vt: [2170, 64] first4: -0.595092 -0.942940 0.130205 0.529570 [Debug] dit_step41_xt: [2170, 64] first4: 0.598236 1.356575 -0.128902 -0.018338 [DiT] Step 42/50 t=0.180 [Debug] dit_step42_vt: [2170, 64] first4: -0.556873 -1.027284 0.163992 0.451921 [Debug] dit_step42_xt: [2170, 64] first4: 0.609373 1.377121 -0.132181 -0.027376 [DiT] Step 43/50 t=0.160 [Debug] dit_step43_vt: [2170, 64] first4: -0.531167 -1.061811 0.225547 0.413319 [Debug] dit_step43_xt: [2170, 64] first4: 0.619996 1.398357 -0.136692 -0.035643 [DiT] Step 44/50 t=0.140 [Debug] dit_step44_vt: [2170, 64] first4: -0.519183 -1.120971 0.260390 0.360472 [Debug] dit_step44_xt: [2170, 64] first4: 0.630380 1.420776 -0.141900 -0.042852 [DiT] Step 45/50 t=0.120 [Debug] dit_step45_vt: [2170, 64] first4: -0.498687 -1.153590 0.329024 0.272266 [Debug] dit_step45_xt: [2170, 64] first4: 0.640354 1.443848 -0.148481 -0.048297 [DiT] Step 46/50 t=0.100 [Debug] dit_step46_vt: [2170, 64] first4: -0.464180 -1.203451 0.317510 0.241646 [Debug] dit_step46_xt: [2170, 64] first4: 0.649638 1.467917 -0.154831 -0.053130 [DiT] Step 47/50 t=0.080 [Debug] dit_step47_vt: [2170, 64] first4: -0.479643 -1.289804 0.308489 0.200017 [Debug] dit_step47_xt: [2170, 64] first4: 0.659230 1.493713 -0.161001 -0.057131 [DiT] Step 48/50 t=0.060 [Debug] dit_step48_vt: [2170, 64] first4: -0.422240 -1.311058 0.311072 0.220052 [Debug] dit_step48_xt: [2170, 64] first4: 0.667675 1.519934 -0.167222 -0.061532 [DiT] Step 49/50 t=0.040 [Debug] dit_step49_vt: [2170, 64] first4: -0.429540 -1.331186 0.256577 0.202842 [Debug] dit_x0: [2170, 64] first4: 0.676266 1.546558 -0.172354 -0.065589 [DiT] Step 50/50 t=0.020 [DiT-Generate] Total: 172594.1 ms (172594.1 ms/sample) [Debug] dit_output: [2170, 64] first4: 0.676266 1.546558 -0.172354 -0.065589 [Store] Unload DiT (1061.2 MB) [GGUF] /mnt/workspace/git/acestep.cpp/models/vae-BF16.gguf: 365 tensors, data at offset 30048 load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] VAE backend: CPU (CPU threads: 16) [VAE] Backend: CPU, Weight buffer: 161.1 MB [VAE] Loaded: 5 blocks, upsample=1920x, F32 activations [Store] Load VAE-Dec: 711 ms [VAE] Tiled decode: 3 tiles (chunk=1024, overlap=64, stride=896) [VAE] Graph: 479 nodes, T_latent=960 [VAE] Upsample factor: 1920.00 (expected ~1920) [VAE] Graph: 479 nodes, T_latent=1024 [VAE] Graph: 479 nodes, T_latent=442 [VAE] Tiled decode done: 3 tiles -> T_audio=4166400 (86.80s @ 48kHz) [VAE-Decode Batch0] Decode: 35007.6 ms [Debug] vae_audio: [2, 4166400] first4: -0.001472 -0.000953 -0.001225 -0.000827 [Store] Unload VAE-Dec (161.1 MB) [MP3] Encoding 86.8s @ 128 kbps, 48000 Hz stereo [MP3] 1388928 bytes (12.0:1), 421 ms (206.05x realtime), 32 threads [MP3] Wrote ggml-sft/request00.mp3 [Ace-Synth] All done 2026-07-12 09:57:15.379 | INFO | acestep.core.generation.handler.init_service_loader:_load_main_model_from_checkpoint:174 - [initialize_service] Attempting to load model with attention implementation: sdpa 2026-07-12 09:57:17.227 | INFO | acestep.core.generation.handler.generate_music:generate_music:304 - [generate_music] Starting generation... 2026-07-12 09:57:17.227 | INFO | acestep.core.generation.handler.generate_music:generate_music:307 - [generate_music] Preparing inputs... 2026-07-12 09:57:17.227 | INFO | acestep.core.generation.handler.generate_music_request:_prepare_reference_and_source_audio:179 - [generate_music] cover task: no src_audio but audio codes provided, proceeding with codes 2026-07-12 09:57:17.230 | INFO | acestep.core.generation.handler.generate_music:_vram_preflight_check:156 - [generate_music] VRAM pre-flight: 81.69 GB free, ~0.94 GB needed (batch=1, duration=88s, mode=turbo). 2026-07-12 09:57:17.236 | INFO | acestep.core.generation.handler.conditioning_target:_prepare_target_latents_and_wavs:78 - [generate_music] Decoding audio codes for item 0... 2026-07-12 09:57:17.242 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_precomputed_lm_hints:31 - [generate_music] Decoding audio codes for LM hints for item 0... 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:119 - ====================================================================== 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:120 - πŸ” [DEBUG] DiT TEXT ENCODER INPUT (Inference) 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:121 - ====================================================================== 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:122 - text_prompt: # Instruction Generate audio semantic tokens based on the given conditions: # Caption An upbeat and anthemic pop-rock track driven by bright, slightly overdriven # Metas - bpm: 83 - timesignature: 4 - keyscale: G major - duration: 88 seconds <|endoftext|> 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:123 - ====================================================================== 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:124 - lyrics_text: # Languages fr # Lyric # Lyric [Intro - Guitar Riff] [Verse 1] Dans le monde des tutos virtuels G ta toise en nouvelle passion Avec Ggendoline et PumbΓ© Γ  midi La communautΓ©, c'est l'unitΓ© Quel joie, une clΓ© [Chorus] Dans le monde des tutos virtuels GΓ’ndoline et PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner [Guitar Solo] [Verse 2] Dans le monde des tutos virtuels GΓ’ndoline, PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner<|endoftext|> 2026-07-12 09:57:17.244 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:125 - ====================================================================== 2026-07-12 09:57:17.247 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:110 - [preprocess_batch] Inferring prompt embeddings... 2026-07-12 09:57:17.256 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:113 - [preprocess_batch] Inferring lyric embeddings... 2026-07-12 09:57:17.256 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:161 - [service_generate] Generating audio... (DiT backend: PyTorch (cuda)) 2026-07-12 09:57:17.262 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:263 - [service_generate] DiT diffusion via PyTorch (cuda)... [Request] Loaded request0.json [Turbo] steps=8, shift=3.0 | acestep-v15-turbo-Q5_K_M.gguf [GGML] Running acestep-v15-turbo-Q5_K_M.gguf... [GGML] Done, 47 dump files [Python] Initializing acestep-v15-turbo... [Python] Generating (acestep-v15-turbo, 8 steps)... Using precomputed LM hints Using precomputed LM hints [Python] Wrote python-turbo/output.wav: 4166400 samples (86.80s @ 48kHz stereo) [Python] Done, 40 dump files [Turbo] Cosine similarities GGML vs Python stage GGML vs Python text_hidden 0.999810 lyric_embed 1.000000 enc_hidden 0.999097 detok_output 0.999843 context 0.999900 noise 1.000000 temb_t 0.999968 hidden_after_proj_in 0.999953 enc_after_cond_emb 0.999197 layer0_sa_output 0.999388 hidden_after_layer0 0.999772 hidden_after_layer6 0.999687 hidden_after_layer12 0.998693 hidden_after_layer18 0.995645 hidden_after_layer23 0.991758 dit_step0_vt 0.969782 dit_step0_xt 0.999933 dit_step1_vt 0.972799 dit_step1_xt 0.999792 dit_step2_vt 0.971332 dit_step2_xt 0.999399 dit_step3_vt 0.974239 dit_step3_xt 0.998577 dit_step4_vt 0.972245 dit_step4_xt 0.996707 dit_step5_vt 0.967987 dit_step5_xt 0.992344 dit_step6_vt 0.963553 dit_step6_xt 0.983770 dit_step7_vt 0.954766 dit_x0 0.970780 vae_audio 0.878433 vae_audio (STFT cosine) 0.969365 [Turbo] Error growth GGML vs Python stage cos max_err mean_err mean_A std_A mean_B std_B dit_step0_xt 0.999933 0.138257 0.007678 -0.002290 0.973047 -0.002342 0.972003 dit_step1_xt 0.999792 0.270737 0.013151 -0.005300 0.942881 -0.005313 0.941730 dit_step2_xt 0.999399 0.460268 0.020988 -0.009308 0.909498 -0.009311 0.908527 dit_step3_xt 0.998577 0.689647 0.031134 -0.014715 0.874229 -0.014577 0.873624 dit_step4_xt 0.996707 0.849085 0.045339 -0.021939 0.842458 -0.021660 0.841995 dit_step5_xt 0.992344 1.317355 0.067062 -0.032318 0.825966 -0.032109 0.824593 dit_step6_xt 0.983770 2.173034 0.100468 -0.046952 0.857183 -0.046482 0.855546 [SFT] steps=50, shift=1.0 | acestep-v15-sft-Q5_K_M.gguf [GGML] Running acestep-v15-sft-Q5_K_M.gguf... [GGML] Done, 131 dump files [Python] Initializing acestep-v15-sft... [Python] Generating (acestep-v15-sft, 50 steps)... Using precomputed LM hints Using precomputed LM hints 0%| | 0/50 [00:00 Text-Enc [Registry] Qwen3-Embedding-0.6B-Q8_0.gguf -> Text-Enc [Registry] acestep-5Hz-lm-0.6B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-0.6B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-4B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q5_K_M.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q6_K.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q8_0.gguf -> LM [Registry] acestep-v15-base-BF16.gguf -> DiT [Registry] acestep-v15-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-base-Q6_K.gguf -> DiT [Registry] acestep-v15-base-Q8_0.gguf -> DiT [Registry] acestep-v15-sft-BF16.gguf -> DiT [Registry] acestep-v15-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-BF16.gguf -> DiT [Registry] acestep-v15-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-continuous-BF16.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift1-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift3-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-base-BF16.gguf -> DiT [Registry] acestep-v15-xl-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-base-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sft-BF16.gguf -> DiT [Registry] acestep-v15-xl-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-turbo-BF16.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q8_0.gguf -> DiT [Registry] scragvae-BF16.gguf -> VAE [Registry] vae-BF16.gguf -> VAE [Store] Created (policy=STRICT) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-turbo-Q5_K_M.gguf: 830 tensors, data at offset 69088 [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-turbo-Q5_K_M.gguf: 830 tensors, data at offset 69088 [Synth-Load] Ready: turbo=yes, fa=no, batch_cfg=yes [Resolve-Params] max audio codes across batch: 434 (86.8s @ 5Hz) [Resolve-T] T=2170, S=1085 [Resolve-T] seed=42, steps=8, guidance=1.0, shift=3.0, duration=88.0s [BPE] Loaded from GGUF: 151643 vocab, 151387 merges 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/acestep.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/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] TextEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/Qwen3-Embedding-0.6B-BF16.gguf: 310 tensors, data at offset 5337568 [Load] TextEncoder: 28L, H=1024, Nh=16/8 [Qwen3] Attn: Q+K+V fused [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 310 tensors, 1136.5 MB into backend [Store] Load TextEnc: 262 ms [Debug] text_hidden: [70, 1024] first4: 3.704526 2.436253 0.222853 -13.131872 [Debug] lyric_embed: [167, 1024] first4: 0.029175 0.032227 -0.022339 -0.028809 [Store] Unload TextEnc (1136.5 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] CondEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-turbo-Q5_K_M.gguf: 830 tensors, data at offset 69088 [Load] LyricEncoder: 8L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [Load] TimbreEncoder: 4L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 141 tensors, 416.2 MB into backend [Load] CondEncoder: lyric(8L), timbre(4L, CLS), text_proj, null_cond [Store] Load CondEnc: 690 ms [CondEnc] Lyric sliding mask: 167x167, window=128 [CondEnc] Timbre sliding mask: 2x2, window=128 (CLS) [Encode] Packed: lyric=167 + timbre=1 + text=70 = 238 tokens [Encode-Text Batch0] 70+167 tokens -> enc_S=238, 120.7 ms [Debug] enc_hidden: [238, 2048] first4: 0.688062 -0.020657 -0.051642 0.022549 [Store] Unload CondEnc (416.2 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] Detokenizer backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-turbo-Q5_K_M.gguf: 830 tensors, data at offset 69088 [WeightCtx] Loaded 30 tensors, 73.2 MB into backend [Load] Detokenizer: FSQ(6->2048) + 2L encoder(S=5, 2048->64) [Store] Load FSQ-Detok: 61 ms [Context] Decoded: 434 codes -> 2170 frames (86.8s @ 25Hz) [Build-Context Batch0] Detokenizer: 420.7 ms, 434 codes [Debug] detok_output: [2170, 64] first4: -0.129311 1.458194 0.298132 -0.651512 [Store] Unload FSQ-Detok (73.2 MB) [Init-Noise Batch0] Philox noise seed=42, [2170, 64] solver=euler [Debug] noise: [2170, 64] first4: 0.194336 2.156250 -0.171875 0.847656 [Debug] context: [2170, 128] first4: -0.129311 1.458194 0.298132 -0.651512 [Init-Noise] Starting: T=2170, S=1085, enc_S=238, steps=8, batch=1 (cover) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] DiT backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-turbo-Q5_K_M.gguf: 830 tensors, data at offset 69088 [DiT] Self-attn: Q+K fused, V separate [DiT] Cross-attn: all separate [DiT] MLP: gate+up fused [Load] null_condition_emb found (CFG available) [WeightCtx] Loaded 630 tensors, 2801.1 MB into backend [Load] DiT: 32 layers, H=2560, Nh=32/8, D=128 [Store] Load DiT: 2032 ms [DiT] Batch N=1, T=2170, S=1085, enc_S=238 [DiT] Graph: 2729 nodes [DiT] Solver: ODE Euler (1 NFE/step, order 1) [Debug] tproj: [15360] first4: 0.007412 0.108993 0.074316 -0.076962 [Debug] temb: [2560] first4: 0.026376 -0.023438 -0.016236 0.047262 [Debug] temb_t: [2560] first4: 0.014536 -0.022293 -0.020594 0.048466 [Debug] temb_r: [2560] first4: 0.011840 -0.001145 0.004359 -0.001204 [Debug] sinusoidal_t: [256] first4: 0.562379 0.789627 0.439928 -0.023645 [Debug] sinusoidal_r: [256] first4: 1.000000 1.000000 1.000000 1.000000 [Debug] temb_lin1_t: [2560] first4: 0.063305 -0.019139 0.138854 -0.115857 [Debug] temb_lin1_r: [2560] first4: -0.008961 0.017832 -0.016914 0.012844 [Debug] hidden_after_proj_in: [2560, 1085] first4: 0.352658 -0.584319 0.235708 -0.182073 [Debug] proj_in_input: [192, 2170] first4: -0.129311 1.458194 0.298132 -0.651512 [Debug] enc_after_cond_emb: [2560, 238] first4: 0.047516 0.666162 0.102776 0.870838 [Debug] layer0_sa_input: [2560, 1085] first4: 0.032149 -0.047514 0.103166 -0.134958 [Debug] layer0_q_after_rope: [128, 32] first4: -0.839122 -2.228545 -2.098012 0.565041 [Debug] layer0_k_after_rope: [128, 8] first4: 1.599450 4.265841 -0.491892 5.391529 [Debug] layer0_sa_output: [2560, 1085] first4: 1.599450 4.265841 -0.491892 5.391529 [Debug] layer0_attn_out: [4096, 1085] first4: -0.839122 -2.228545 -2.098012 0.565041 [Debug] layer0_after_self_attn: [2560, 1085] first4: 0.478263 -2.237561 -0.125057 -2.763805 [Debug] layer0_after_cross_attn: [2560, 1085] first4: 0.361870 -1.976914 0.028454 -3.570030 [Debug] hidden_after_layer0: [2560, 1085] first4: 2.114879 -2.339164 233.534317 -3.539886 [Debug] hidden_after_layer6: [2560, 1085] first4: -6.125286 -2.177757 308.441315 -12.089705 [Debug] hidden_after_layer12: [2560, 1085] first4: -12.196932 29.472122 251.589767 28.293297 [Debug] hidden_after_layer18: [2560, 1085] first4: 43.714672 12.571980 253.607864 5.532092 [Debug] hidden_after_layer31: [2560, 1085] first4: -33.216412 -231.121124 -161.159271 156.747391 [Debug] dit_step0_vt: [2170, 64] first4: -1.126410 1.704001 0.905653 1.397307 [Debug] dit_step0_xt: [2170, 64] first4: 0.245536 2.078795 -0.213041 0.784142 [DiT] Step 1/8 t=1.000 [Debug] dit_step1_vt: [2170, 64] first4: -0.750821 2.483658 0.810392 1.135372 [Debug] dit_step1_xt: [2170, 64] first4: 0.286490 1.943323 -0.257244 0.722213 [DiT] Step 2/8 t=0.955 [Debug] dit_step2_vt: [2170, 64] first4: -0.906925 1.360082 1.107049 1.062083 [Debug] dit_step2_xt: [2170, 64] first4: 0.346952 1.852651 -0.331048 0.651407 [DiT] Step 3/8 t=0.900 [Debug] dit_step3_vt: [2170, 64] first4: -0.385491 1.384823 0.952933 0.944730 [Debug] dit_step3_xt: [2170, 64] first4: 0.379076 1.737249 -0.410459 0.572680 [DiT] Step 4/8 t=0.833 [Debug] dit_step4_vt: [2170, 64] first4: -0.515544 1.460281 1.086141 0.878154 [Debug] dit_step4_xt: [2170, 64] first4: 0.434313 1.580790 -0.526831 0.478592 [DiT] Step 5/8 t=0.750 [Debug] dit_step5_vt: [2170, 64] first4: 0.311581 2.482050 0.781654 0.754865 [Debug] dit_step5_xt: [2170, 64] first4: 0.389802 1.226212 -0.638496 0.370754 [DiT] Step 6/8 t=0.643 [Debug] dit_step6_vt: [2170, 64] first4: 0.516061 2.131855 0.497089 0.391451 [Debug] dit_step6_xt: [2170, 64] first4: 0.286589 0.799841 -0.737913 0.292464 [DiT] Step 7/8 t=0.500 [Debug] dit_step7_vt: [2170, 64] first4: -1.462179 2.606373 1.243745 0.152383 [Debug] dit_x0: [2170, 64] first4: 0.725243 0.017929 -1.111037 0.246749 [DiT] Step 8/8 t=0.300 [DiT-Generate] Total: 73549.4 ms (73549.4 ms/sample) [Debug] dit_output: [2170, 64] first4: 0.725243 0.017929 -1.111037 0.246749 [Store] Unload DiT (2801.1 MB) [GGUF] /mnt/workspace/git/acestep.cpp/models/vae-BF16.gguf: 365 tensors, data at offset 30048 load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] VAE backend: CPU (CPU threads: 16) [VAE] Backend: CPU, Weight buffer: 161.1 MB [VAE] Loaded: 5 blocks, upsample=1920x, F32 activations [Store] Load VAE-Dec: 718 ms [VAE] Tiled decode: 3 tiles (chunk=1024, overlap=64, stride=896) [VAE] Graph: 479 nodes, T_latent=960 [VAE] Upsample factor: 1920.00 (expected ~1920) [VAE] Graph: 479 nodes, T_latent=1024 [VAE] Graph: 479 nodes, T_latent=442 [VAE] Tiled decode done: 3 tiles -> T_audio=4166400 (86.80s @ 48kHz) [VAE-Decode Batch0] Decode: 34700.2 ms [Debug] vae_audio: [2, 4166400] first4: 0.001464 0.003523 0.003501 0.003182 [Store] Unload VAE-Dec (161.1 MB) [MP3] Encoding 86.8s @ 128 kbps, 48000 Hz stereo [MP3] 1388928 bytes (12.0:1), 519 ms (167.30x realtime), 32 threads [MP3] Wrote ggml-xl-turbo/request00.mp3 [Ace-Synth] All done 2026-07-12 09:59:13.908 | INFO | acestep.core.generation.handler.init_service_loader:_load_main_model_from_checkpoint:174 - [initialize_service] Attempting to load model with attention implementation: sdpa [Python] Wrote python-sft/output.wav: 4166400 samples (86.80s @ 48kHz stereo) [Python] Done, 166 dump files [SFT] Cosine similarities GGML vs Python stage GGML vs Python text_hidden 0.999810 lyric_embed 1.000000 enc_hidden 0.999097 detok_output 0.999843 context 0.999900 noise 1.000000 temb_t 0.999877 hidden_after_proj_in 0.999954 enc_after_cond_emb 0.999193 layer0_sa_output 0.999445 hidden_after_layer0 0.999822 hidden_after_layer6 0.999563 hidden_after_layer12 0.998939 hidden_after_layer18 0.997982 hidden_after_layer23 0.998429 dit_step0_vt 0.998342 dit_step0_xt 1.000000 dit_step5_vt 0.998480 dit_step5_xt 0.999965 dit_step10_vt 0.997784 dit_step10_xt 0.999876 dit_step15_vt 0.995926 dit_step15_xt 0.999668 dit_step20_vt 0.992381 dit_step20_xt 0.999157 dit_step25_vt 0.987022 dit_step25_xt 0.998153 dit_step30_vt 0.980277 dit_step30_xt 0.996442 dit_step35_vt 0.973241 dit_step35_xt 0.994201 dit_step40_vt 0.969022 dit_step40_xt 0.991906 dit_step45_vt 0.973988 dit_step45_xt 0.990269 dit_step49_vt 0.981958 dit_x0 0.989708 vae_audio 0.953278 vae_audio (STFT cosine) 0.979799 [SFT] Error growth GGML vs Python stage cos max_err mean_err mean_A std_A mean_B std_B dit_step0_xt 1.000000 0.019056 0.001597 -0.001648 0.978528 -0.001619 0.978872 dit_step5_xt 0.999965 0.068526 0.005859 -0.008321 0.885208 -0.008480 0.883811 dit_step10_xt 0.999876 0.093556 0.009715 -0.015148 0.801492 -0.015075 0.802324 dit_step15_xt 0.999668 0.156803 0.014320 -0.022021 0.731169 -0.021933 0.731304 dit_step20_xt 0.999157 0.372291 0.020689 -0.028992 0.678592 -0.028882 0.679066 dit_step25_xt 0.998153 0.632104 0.028646 -0.036076 0.648018 -0.035643 0.650780 dit_step30_xt 0.996442 0.906749 0.038914 -0.043192 0.642720 -0.042577 0.647107 dit_step35_xt 0.994201 1.201091 0.050999 -0.050286 0.663605 -0.049471 0.669256 dit_step40_xt 0.991906 1.481584 0.064399 -0.057383 0.708304 -0.056532 0.715783 dit_step45_xt 0.990269 1.721631 0.077110 -0.064513 0.772575 -0.063533 0.780575 [Xl-turbo] steps=8, shift=3.0 | acestep-v15-xl-turbo-Q5_K_M.gguf [GGML] Running acestep-v15-xl-turbo-Q5_K_M.gguf... [GGML] Done, 47 dump files [Python] Initializing acestep-v15-xl-turbo... Loading checkpoint shards: 0%| | 0/4 [00:00 1.0 (turbo does not use CFG). 2026-07-12 09:59:16.517 | INFO | acestep.core.generation.handler.generate_music:generate_music:304 - [generate_music] Starting generation... 2026-07-12 09:59:16.517 | INFO | acestep.core.generation.handler.generate_music:generate_music:307 - [generate_music] Preparing inputs... 2026-07-12 09:59:16.518 | INFO | acestep.core.generation.handler.generate_music_request:_prepare_reference_and_source_audio:179 - [generate_music] cover task: no src_audio but audio codes provided, proceeding with codes 2026-07-12 09:59:16.521 | INFO | acestep.core.generation.handler.generate_music:_vram_preflight_check:156 - [generate_music] VRAM pre-flight: 76.75 GB free, ~1.23 GB needed (batch=1, duration=88s, mode=xl_turbo). 2026-07-12 09:59:16.525 | INFO | acestep.core.generation.handler.conditioning_target:_prepare_target_latents_and_wavs:78 - [generate_music] Decoding audio codes for item 0... 2026-07-12 09:59:16.530 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_precomputed_lm_hints:31 - [generate_music] Decoding audio codes for LM hints for item 0... 2026-07-12 09:59:16.531 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:119 - ====================================================================== 2026-07-12 09:59:16.532 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:120 - πŸ” [DEBUG] DiT TEXT ENCODER INPUT (Inference) 2026-07-12 09:59:16.532 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:121 - ====================================================================== 2026-07-12 09:59:16.532 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:122 - text_prompt: # Instruction Generate audio semantic tokens based on the given conditions: # Caption An upbeat and anthemic pop-rock track driven by bright, slightly overdriven # Metas - bpm: 83 - timesignature: 4 - keyscale: G major - duration: 88 seconds <|endoftext|> 2026-07-12 09:59:16.532 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:123 - ====================================================================== 2026-07-12 09:59:16.532 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:124 - lyrics_text: # Languages fr # Lyric # Lyric [Intro - Guitar Riff] [Verse 1] Dans le monde des tutos virtuels G ta toise en nouvelle passion Avec Ggendoline et PumbΓ© Γ  midi La communautΓ©, c'est l'unitΓ© Quel joie, une clΓ© [Chorus] Dans le monde des tutos virtuels GΓ’ndoline et PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner [Guitar Solo] [Verse 2] Dans le monde des tutos virtuels GΓ’ndoline, PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner<|endoftext|> 2026-07-12 09:59:16.532 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:125 - ====================================================================== 2026-07-12 09:59:16.535 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:110 - [preprocess_batch] Inferring prompt embeddings... 2026-07-12 09:59:16.543 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:113 - [preprocess_batch] Inferring lyric embeddings... 2026-07-12 09:59:16.543 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:161 - [service_generate] Generating audio... (DiT backend: PyTorch (cuda)) 2026-07-12 09:59:16.550 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:263 - [service_generate] DiT diffusion via PyTorch (cuda)... 2026-07-12 09:59:16.556 | INFO | acestep.models.common.dcw_correction:__init__:98 - [DCW] Active β€” mode=double, scaler=0.0500, high_scaler=0.0200, wavelet='haar' 2026-07-12 09:59:17.102 | INFO | acestep.core.generation.handler.generate_music_decode:_prepare_generate_music_decode_state:42 - [generate_music] Model generation completed. Decoding latents... 2026-07-12 09:59:17.103 | DEBUG | acestep.core.generation.handler.generate_music_decode:_prepare_generate_music_decode_state:64 - [generate_music] pred_latents: torch.Size([1, 2170, 64]), dtype=torch.bfloat16 2026-07-12 09:59:17.103 | DEBUG | acestep.core.generation.handler.generate_music_decode:_prepare_generate_music_decode_state:65 - [generate_music] time_costs: {'encoder_time_cost': 0.006358623504638672, 'diffusion_time_cost': 0.5455818176269531, 'diffusion_per_step_time_cost': 0.06819772720336914, 'total_time_cost': 0.5519404411315918, 'offload_time_cost': 0.0} 2026-07-12 09:59:17.126 | INFO | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:127 - [generate_music] Decoding latents with VAE... 2026-07-12 09:59:17.127 | DEBUG | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:136 - [generate_music] Before VAE decode: allocated=17.03GB, max=18.34GB 2026-07-12 09:59:17.127 | INFO | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:154 - [generate_music] Effective free VRAM before VAE decode: 76.66 GB 2026-07-12 09:59:17.127 | INFO | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:172 - [generate_music] Using tiled VAE decode to reduce VRAM usage... 2026-07-12 09:59:17.127 | DEBUG | acestep.core.generation.handler.memory_utils:_get_auto_decode_chunk_size:108 - [_get_auto_decode_chunk_size] Effective free VRAM: 76.66 GB 2026-07-12 09:59:17.127 | DEBUG | acestep.core.generation.handler.memory_utils:_should_offload_wav_to_cpu:131 - [_should_offload_wav_to_cpu] Effective free VRAM: 76.66 GB 2026-07-12 09:59:17.127 | INFO | acestep.core.generation.handler.vae_decode:tiled_decode:56 - [tiled_decode] chunk_size=512, offload_wav_to_cpu=False, latents_shape=torch.Size([1, 64, 2170]) 2026-07-12 09:59:17.398 | DEBUG | acestep.core.generation.handler.generate_music_decode:_decode_generate_music_pred_latents:193 - [generate_music] After VAE decode: allocated=17.21GB, max=18.50GB 2026-07-12 09:59:17.606 | INFO | acestep.core.generation.handler.generate_music_payload:_build_generate_music_success_payload:38 - [generate_music] VAE decode completed. Preparing audio tensors... 2026-07-12 09:59:17.610 | INFO | acestep.core.generation.handler.generate_music_payload:_build_generate_music_success_payload:50 - [generate_music] Done! Generated 1 audio tensors. [Request] Parsed ggml-xl-sft/request0.json [Request] seed=42 lm_batch=1 synth_batch=1 [Request] caption: An upbeat and anthemic pop-rock track driven by bright, slig... [Request] lyrics: 490 bytes [Request] bpm=83 dur=88 key=G major ts=4 lang=fr [Request] lm: temp=0.85 cfg=2.0 top_p=0.90 top_k=0 [Request] dit: steps=50 guidance=1.0 shift=1.0 [Request] task_type: text2music [Request] solver: euler (stork_substeps=10) [Request] lm_mode: generate [Request] output_format: mp3 [Request] synth_model: acestep-v15-xl-sft-Q5_K_M.gguf [Request] vae: vae-BF16.gguf [Request] audio_codes: (present) [Ace-Synth] Batch: 1 request(s) [Registry] Qwen3-Embedding-0.6B-BF16.gguf -> Text-Enc [Registry] Qwen3-Embedding-0.6B-Q8_0.gguf -> Text-Enc [Registry] acestep-5Hz-lm-0.6B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-0.6B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-1.7B-Q8_0.gguf -> LM [Registry] acestep-5Hz-lm-4B-BF16.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q5_K_M.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q6_K.gguf -> LM [Registry] acestep-5Hz-lm-4B-Q8_0.gguf -> LM [Registry] acestep-v15-base-BF16.gguf -> DiT [Registry] acestep-v15-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-base-Q6_K.gguf -> DiT [Registry] acestep-v15-base-Q8_0.gguf -> DiT [Registry] acestep-v15-sft-BF16.gguf -> DiT [Registry] acestep-v15-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-BF16.gguf -> DiT [Registry] acestep-v15-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-continuous-BF16.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-continuous-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift1-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift1-Q8_0.gguf -> DiT [Registry] acestep-v15-turbo-shift3-BF16.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q4_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q5_K_M.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q6_K.gguf -> DiT [Registry] acestep-v15-turbo-shift3-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-base-BF16.gguf -> DiT [Registry] acestep-v15-xl-base-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-base-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-base-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sft-BF16.gguf -> DiT [Registry] acestep-v15-xl-sft-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sft-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sft-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-BF16.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-sftturbo50-Q8_0.gguf -> DiT [Registry] acestep-v15-xl-turbo-BF16.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q4_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q5_K_M.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q6_K.gguf -> DiT [Registry] acestep-v15-xl-turbo-Q8_0.gguf -> DiT [Registry] scragvae-BF16.gguf -> VAE [Registry] vae-BF16.gguf -> VAE [Store] Created (policy=STRICT) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-sft-Q5_K_M.gguf: 830 tensors, data at offset 69056 [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-sft-Q5_K_M.gguf: 830 tensors, data at offset 69056 [Synth-Load] Ready: turbo=no, fa=no, batch_cfg=yes [Resolve-Params] max audio codes across batch: 434 (86.8s @ 5Hz) [Resolve-T] T=2170, S=1085 [Resolve-T] seed=42, steps=50, guidance=1.0, shift=1.0, duration=88.0s [BPE] Loaded from GGUF: 151643 vocab, 151387 merges 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/acestep.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/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] TextEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/Qwen3-Embedding-0.6B-BF16.gguf: 310 tensors, data at offset 5337568 [Load] TextEncoder: 28L, H=1024, Nh=16/8 [Qwen3] Attn: Q+K+V fused [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 310 tensors, 1136.5 MB into backend [Store] Load TextEnc: 468 ms [Debug] text_hidden: [70, 1024] first4: 3.704526 2.436253 0.222853 -13.131872 [Debug] lyric_embed: [167, 1024] first4: 0.029175 0.032227 -0.022339 -0.028809 [Store] Unload TextEnc (1136.5 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] CondEncoder backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-sft-Q5_K_M.gguf: 830 tensors, data at offset 69056 [Load] LyricEncoder: 8L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [Load] TimbreEncoder: 4L [Qwen3] Attn: Q+K fused, V separate [Qwen3] MLP: gate+up fused [WeightCtx] Loaded 141 tensors, 416.2 MB into backend [Load] CondEncoder: lyric(8L), timbre(4L, CLS), text_proj, null_cond [Store] Load CondEnc: 298 ms [CondEnc] Lyric sliding mask: 167x167, window=128 [CondEnc] Timbre sliding mask: 2x2, window=128 (CLS) [Encode] Packed: lyric=167 + timbre=1 + text=70 = 238 tokens [Encode-Text Batch0] 70+167 tokens -> enc_S=238, 120.2 ms [Debug] enc_hidden: [238, 2048] first4: 0.688062 -0.020657 -0.051642 0.022549 [Store] Unload CondEnc (416.2 MB) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] Detokenizer backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-sft-Q5_K_M.gguf: 830 tensors, data at offset 69056 [WeightCtx] Loaded 30 tensors, 73.2 MB into backend [Load] Detokenizer: FSQ(6->2048) + 2L encoder(S=5, 2048->64) [Store] Load FSQ-Detok: 65 ms [Context] Decoded: 434 codes -> 2170 frames (86.8s @ 25Hz) [Build-Context Batch0] Detokenizer: 423.4 ms, 434 codes [Debug] detok_output: [2170, 64] first4: -0.129311 1.458194 0.298132 -0.651512 [Store] Unload FSQ-Detok (73.2 MB) [Init-Noise Batch0] Philox noise seed=42, [2170, 64] solver=euler [Debug] noise: [2170, 64] first4: 0.194336 2.156250 -0.171875 0.847656 [Debug] context: [2170, 128] first4: -0.129311 1.458194 0.298132 -0.651512 [Init-Noise] Starting: T=2170, S=1085, enc_S=238, steps=50, batch=1 (cover) load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] DiT backend: CPU (CPU threads: 16) [GGUF] /mnt/workspace/git/acestep.cpp/models/acestep-v15-xl-sft-Q5_K_M.gguf: 830 tensors, data at offset 69056 [DiT] Self-attn: Q+K fused, V separate [DiT] Cross-attn: all separate [DiT] MLP: gate+up fused [Load] null_condition_emb found (CFG available) [WeightCtx] Loaded 630 tensors, 2801.1 MB into backend [Load] DiT: 32 layers, H=2560, Nh=32/8, D=128 [Store] Load DiT: 2102 ms [DiT] Batch N=1, T=2170, S=1085, enc_S=238 [DiT] Graph: 2729 nodes [DiT] Solver: ODE Euler (1 NFE/step, order 1) [Debug] tproj: [15360] first4: 0.004454 0.088804 0.077117 -0.035327 [Debug] temb: [2560] first4: 0.010106 -0.007940 -0.006171 0.004931 [Debug] temb_t: [2560] first4: 0.008579 -0.008000 -0.006857 0.003932 [Debug] temb_r: [2560] first4: 0.001526 0.000060 0.000686 0.000999 [Debug] sinusoidal_t: [256] first4: 0.562379 0.789627 0.439928 -0.023645 [Debug] sinusoidal_r: [256] first4: 1.000000 1.000000 1.000000 1.000000 [Debug] temb_lin1_t: [2560] first4: 0.028611 -0.079612 0.086698 -0.070769 [Debug] temb_lin1_r: [2560] first4: 0.006506 -0.006323 -0.000918 -0.013460 [Debug] hidden_after_proj_in: [2560, 1085] first4: 0.342082 -0.499954 0.107260 -0.205408 [Debug] proj_in_input: [192, 2170] first4: -0.129311 1.458194 0.298132 -0.651512 [Debug] enc_after_cond_emb: [2560, 238] first4: 0.058434 0.658213 0.110100 0.865776 [Debug] layer0_sa_input: [2560, 1085] first4: 0.029444 -0.037461 0.102438 -0.098226 [Debug] layer0_q_after_rope: [128, 32] first4: -0.869956 -2.039738 -2.888869 0.733576 [Debug] layer0_k_after_rope: [128, 8] first4: 1.885306 4.325730 -0.780136 7.750941 [Debug] layer0_sa_output: [2560, 1085] first4: 1.885306 4.325730 -0.780136 7.750941 [Debug] layer0_attn_out: [4096, 1085] first4: -0.869956 -2.039738 -2.888869 0.733576 [Debug] layer0_after_self_attn: [2560, 1085] first4: 0.461121 -2.160442 -0.468358 -4.108160 [Debug] layer0_after_cross_attn: [2560, 1085] first4: 0.295444 -1.889656 -0.392346 -4.112534 [Debug] hidden_after_layer0: [2560, 1085] first4: -1.380714 -1.249189 240.959122 -3.969363 [Debug] hidden_after_layer6: [2560, 1085] first4: -7.057535 3.304931 301.314941 -21.764675 [Debug] hidden_after_layer12: [2560, 1085] first4: -21.769068 37.141205 219.646042 14.473962 [Debug] hidden_after_layer18: [2560, 1085] first4: 12.172172 -14.774710 226.105438 20.880741 [Debug] hidden_after_layer31: [2560, 1085] first4: 64.596344 -116.611908 -3.008900 78.931442 [Debug] dit_step0_vt: [2170, 64] first4: -0.346020 2.224401 0.054148 1.693466 [Debug] dit_step0_xt: [2170, 64] first4: 0.201256 2.111762 -0.172958 0.813787 [DiT] Step 1/50 t=1.000 [Debug] dit_step1_vt: [2170, 64] first4: -0.367564 2.193091 0.096830 1.562009 [Debug] dit_step1_xt: [2170, 64] first4: 0.208608 2.067900 -0.174895 0.782547 [DiT] Step 2/50 t=0.980 [Debug] dit_step2_vt: [2170, 64] first4: -0.348577 2.168818 0.082129 1.537476 [Debug] dit_step2_xt: [2170, 64] first4: 0.215579 2.024524 -0.176537 0.751797 [DiT] Step 3/50 t=0.960 [Debug] dit_step3_vt: [2170, 64] first4: -0.313320 2.119481 0.071088 1.576522 [Debug] dit_step3_xt: [2170, 64] first4: 0.221846 1.982134 -0.177959 0.720267 [DiT] Step 4/50 t=0.940 [Debug] dit_step4_vt: [2170, 64] first4: -0.295707 2.064621 0.070415 1.569607 [Debug] dit_step4_xt: [2170, 64] first4: 0.227760 1.940842 -0.179367 0.688875 [DiT] Step 5/50 t=0.920 [Debug] dit_step5_vt: [2170, 64] first4: -0.280971 2.037744 0.100077 1.573523 [Debug] dit_step5_xt: [2170, 64] first4: 0.233379 1.900087 -0.181369 0.657404 [DiT] Step 6/50 t=0.900 [Debug] dit_step6_vt: [2170, 64] first4: -0.268591 2.041671 0.133152 1.572915 [Debug] dit_step6_xt: [2170, 64] first4: 0.238751 1.859254 -0.184032 0.625946 [DiT] Step 7/50 t=0.880 [Debug] dit_step7_vt: [2170, 64] first4: -0.274806 1.995407 0.136987 1.590401 [Debug] dit_step7_xt: [2170, 64] first4: 0.244247 1.819346 -0.186772 0.594138 [DiT] Step 8/50 t=0.860 [Debug] dit_step8_vt: [2170, 64] first4: -0.299756 1.925648 0.194119 1.585199 [Debug] dit_step8_xt: [2170, 64] first4: 0.250242 1.780833 -0.190654 0.562434 [DiT] Step 9/50 t=0.840 [Debug] dit_step9_vt: [2170, 64] first4: -0.339567 1.854802 0.166882 1.569588 [Debug] dit_step9_xt: [2170, 64] first4: 0.257033 1.743737 -0.193992 0.531042 [DiT] Step 10/50 t=0.820 [Debug] dit_step10_vt: [2170, 64] first4: -0.327531 1.806104 0.162155 1.565610 [Debug] dit_step10_xt: [2170, 64] first4: 0.263584 1.707614 -0.197235 0.499730 [DiT] Step 11/50 t=0.800 [Debug] dit_step11_vt: [2170, 64] first4: -0.327983 1.743746 0.128584 1.553848 [Debug] dit_step11_xt: [2170, 64] first4: 0.270144 1.672740 -0.199806 0.468653 [DiT] Step 12/50 t=0.780 [Debug] dit_step12_vt: [2170, 64] first4: -0.337777 1.714856 0.135273 1.505397 [Debug] dit_step12_xt: [2170, 64] first4: 0.276899 1.638442 -0.202512 0.438545 [DiT] Step 13/50 t=0.760 [Debug] dit_step13_vt: [2170, 64] first4: -0.319448 1.669666 0.097199 1.500035 [Debug] dit_step13_xt: [2170, 64] first4: 0.283288 1.605049 -0.204456 0.408544 [DiT] Step 14/50 t=0.740 [Debug] dit_step14_vt: [2170, 64] first4: -0.310587 1.626640 0.064327 1.528694 [Debug] dit_step14_xt: [2170, 64] first4: 0.289500 1.572516 -0.205742 0.377970 [DiT] Step 15/50 t=0.720 [Debug] dit_step15_vt: [2170, 64] first4: -0.311207 1.564853 0.067035 1.485400 [Debug] dit_step15_xt: [2170, 64] first4: 0.295724 1.541219 -0.207083 0.348263 [DiT] Step 16/50 t=0.700 [Debug] dit_step16_vt: [2170, 64] first4: -0.286951 1.512848 0.050197 1.516992 [Debug] dit_step16_xt: [2170, 64] first4: 0.301463 1.510962 -0.208087 0.317923 [DiT] Step 17/50 t=0.680 [Debug] dit_step17_vt: [2170, 64] first4: -0.236430 1.469312 0.015441 1.491553 [Debug] dit_step17_xt: [2170, 64] first4: 0.306192 1.481576 -0.208396 0.288092 [DiT] Step 18/50 t=0.660 [Debug] dit_step18_vt: [2170, 64] first4: -0.203499 1.413997 -0.003282 1.496484 [Debug] dit_step18_xt: [2170, 64] first4: 0.310262 1.453296 -0.208330 0.258162 [DiT] Step 19/50 t=0.640 [Debug] dit_step19_vt: [2170, 64] first4: -0.160276 1.345931 -0.005655 1.475027 [Debug] dit_step19_xt: [2170, 64] first4: 0.313467 1.426378 -0.208217 0.228661 [DiT] Step 20/50 t=0.620 [Debug] dit_step20_vt: [2170, 64] first4: -0.121050 1.333074 -0.049496 1.451948 [Debug] dit_step20_xt: [2170, 64] first4: 0.315888 1.399716 -0.207227 0.199623 [DiT] Step 21/50 t=0.600 [Debug] dit_step21_vt: [2170, 64] first4: -0.097318 1.302594 -0.054393 1.485944 [Debug] dit_step21_xt: [2170, 64] first4: 0.317835 1.373664 -0.206139 0.169904 [DiT] Step 22/50 t=0.580 [Debug] dit_step22_vt: [2170, 64] first4: -0.043720 1.241534 -0.059537 1.437499 [Debug] dit_step22_xt: [2170, 64] first4: 0.318709 1.348834 -0.204948 0.141154 [DiT] Step 23/50 t=0.560 [Debug] dit_step23_vt: [2170, 64] first4: -0.076813 1.191223 -0.045373 1.386676 [Debug] dit_step23_xt: [2170, 64] first4: 0.320245 1.325009 -0.204041 0.113420 [DiT] Step 24/50 t=0.540 [Debug] dit_step24_vt: [2170, 64] first4: -0.069844 1.167923 -0.064522 1.403782 [Debug] dit_step24_xt: [2170, 64] first4: 0.321642 1.301651 -0.202751 0.085344 [DiT] Step 25/50 t=0.520 [Debug] dit_step25_vt: [2170, 64] first4: -0.026637 1.056408 -0.058297 1.395429 [Debug] dit_step25_xt: [2170, 64] first4: 0.322175 1.280523 -0.201585 0.057436 [DiT] Step 26/50 t=0.500 [Debug] dit_step26_vt: [2170, 64] first4: 0.014232 1.032969 -0.078819 1.367834 [Debug] dit_step26_xt: [2170, 64] first4: 0.321890 1.259863 -0.200008 0.030079 [DiT] Step 27/50 t=0.480 [Debug] dit_step27_vt: [2170, 64] first4: -0.017686 0.983749 -0.115745 1.330849 [Debug] dit_step27_xt: [2170, 64] first4: 0.322244 1.240188 -0.197693 0.003462 [DiT] Step 28/50 t=0.460 [Debug] dit_step28_vt: [2170, 64] first4: -0.023561 0.919694 -0.109643 1.309263 [Debug] dit_step28_xt: [2170, 64] first4: 0.322715 1.221794 -0.195500 -0.022723 [DiT] Step 29/50 t=0.440 [Debug] dit_step29_vt: [2170, 64] first4: -0.069057 0.824366 -0.125542 1.311419 [Debug] dit_step29_xt: [2170, 64] first4: 0.324096 1.205307 -0.192990 -0.048951 [DiT] Step 30/50 t=0.420 [Debug] dit_step30_vt: [2170, 64] first4: -0.063393 0.784541 -0.103994 1.270113 [Debug] dit_step30_xt: [2170, 64] first4: 0.325364 1.189616 -0.190910 -0.074354 [DiT] Step 31/50 t=0.400 [Debug] dit_step31_vt: [2170, 64] first4: -0.062742 0.680851 -0.125309 1.211003 [Debug] dit_step31_xt: [2170, 64] first4: 0.326619 1.175999 -0.188404 -0.098574 [DiT] Step 32/50 t=0.380 [Debug] dit_step32_vt: [2170, 64] first4: -0.047293 0.618241 -0.142050 1.172474 [Debug] dit_step32_xt: [2170, 64] first4: 0.327565 1.163634 -0.185563 -0.122023 [DiT] Step 33/50 t=0.360 [Debug] dit_step33_vt: [2170, 64] first4: -0.065793 0.510059 -0.153331 1.178169 [Debug] dit_step33_xt: [2170, 64] first4: 0.328881 1.153433 -0.182496 -0.145587 [DiT] Step 34/50 t=0.340 [Debug] dit_step34_vt: [2170, 64] first4: -0.094661 0.442633 -0.158607 1.128117 [Debug] dit_step34_xt: [2170, 64] first4: 0.330774 1.144580 -0.179324 -0.168149 [DiT] Step 35/50 t=0.320 [Debug] dit_step35_vt: [2170, 64] first4: -0.020365 0.361930 -0.175097 1.072740 [Debug] dit_step35_xt: [2170, 64] first4: 0.331181 1.137342 -0.175822 -0.189604 [DiT] Step 36/50 t=0.300 [Debug] dit_step36_vt: [2170, 64] first4: -0.103992 0.278609 -0.136187 1.045243 [Debug] dit_step36_xt: [2170, 64] first4: 0.333261 1.131770 -0.173098 -0.210509 [DiT] Step 37/50 t=0.280 [Debug] dit_step37_vt: [2170, 64] first4: -0.097077 0.145383 -0.103258 1.059899 [Debug] dit_step37_xt: [2170, 64] first4: 0.335203 1.128862 -0.171033 -0.231707 [DiT] Step 38/50 t=0.260 [Debug] dit_step38_vt: [2170, 64] first4: -0.133811 0.017991 -0.102795 0.998967 [Debug] dit_step38_xt: [2170, 64] first4: 0.337879 1.128502 -0.168977 -0.251686 [DiT] Step 39/50 t=0.240 [Debug] dit_step39_vt: [2170, 64] first4: -0.169362 -0.103795 -0.128939 0.931427 [Debug] dit_step39_xt: [2170, 64] first4: 0.341266 1.130578 -0.166398 -0.270314 [DiT] Step 40/50 t=0.220 [Debug] dit_step40_vt: [2170, 64] first4: -0.184650 -0.165024 -0.105861 0.883659 [Debug] dit_step40_xt: [2170, 64] first4: 0.344959 1.133878 -0.164281 -0.287988 [DiT] Step 41/50 t=0.200 [Debug] dit_step41_vt: [2170, 64] first4: -0.214039 -0.318036 -0.083044 0.842932 [Debug] dit_step41_xt: [2170, 64] first4: 0.349240 1.140239 -0.162620 -0.304846 [DiT] Step 42/50 t=0.180 [Debug] dit_step42_vt: [2170, 64] first4: -0.228750 -0.399931 -0.069324 0.799776 [Debug] dit_step42_xt: [2170, 64] first4: 0.353815 1.148238 -0.161234 -0.320842 [DiT] Step 43/50 t=0.160 [Debug] dit_step43_vt: [2170, 64] first4: -0.276675 -0.492975 -0.056593 0.745258 [Debug] dit_step43_xt: [2170, 64] first4: 0.359348 1.158097 -0.160102 -0.335747 [DiT] Step 44/50 t=0.140 [Debug] dit_step44_vt: [2170, 64] first4: -0.285736 -0.579473 0.030705 0.766952 [Debug] dit_step44_xt: [2170, 64] first4: 0.365063 1.169687 -0.160716 -0.351086 [DiT] Step 45/50 t=0.120 [Debug] dit_step45_vt: [2170, 64] first4: -0.275148 -0.635852 0.067623 0.648227 [Debug] dit_step45_xt: [2170, 64] first4: 0.370566 1.182404 -0.162068 -0.364051 [DiT] Step 46/50 t=0.100 [Debug] dit_step46_vt: [2170, 64] first4: -0.323181 -0.695716 0.060004 0.590187 [Debug] dit_step46_xt: [2170, 64] first4: 0.377030 1.196318 -0.163269 -0.375854 [DiT] Step 47/50 t=0.080 [Debug] dit_step47_vt: [2170, 64] first4: -0.380049 -0.769904 0.161810 0.576089 [Debug] dit_step47_xt: [2170, 64] first4: 0.384631 1.211716 -0.166505 -0.387376 [DiT] Step 48/50 t=0.060 [Debug] dit_step48_vt: [2170, 64] first4: -0.419213 -0.828707 0.260861 0.528233 [Debug] dit_step48_xt: [2170, 64] first4: 0.393015 1.228290 -0.171722 -0.397941 [DiT] Step 49/50 t=0.040 [Debug] dit_step49_vt: [2170, 64] first4: -0.417198 -0.836585 0.193649 0.546110 [Debug] dit_x0: [2170, 64] first4: 0.401359 1.245022 -0.175595 -0.408863 [DiT] Step 50/50 t=0.020 [DiT-Generate] Total: 464499.1 ms (464499.1 ms/sample) [Debug] dit_output: [2170, 64] first4: 0.401359 1.245022 -0.175595 -0.408863 [Store] Unload DiT (2801.1 MB) [GGUF] /mnt/workspace/git/acestep.cpp/models/vae-BF16.gguf: 365 tensors, data at offset 30048 load_backend: loaded CUDA backend from /mnt/workspace/git/acestep.cpp/build/libggml-cuda.so load_backend: loaded Vulkan backend from /mnt/workspace/git/acestep.cpp/build/libggml-vulkan.so load_backend: loaded CPU backend from /mnt/workspace/git/acestep.cpp/build/libggml-cpu-zen4.so [Load] VAE backend: CPU (CPU threads: 16) [VAE] Backend: CPU, Weight buffer: 161.1 MB [VAE] Loaded: 5 blocks, upsample=1920x, F32 activations [Store] Load VAE-Dec: 734 ms [VAE] Tiled decode: 3 tiles (chunk=1024, overlap=64, stride=896) [VAE] Graph: 479 nodes, T_latent=960 [VAE] Upsample factor: 1920.00 (expected ~1920) [VAE] Graph: 479 nodes, T_latent=1024 [VAE] Graph: 479 nodes, T_latent=442 [VAE] Tiled decode done: 3 tiles -> T_audio=4166400 (86.80s @ 48kHz) [VAE-Decode Batch0] Decode: 36764.5 ms [Debug] vae_audio: [2, 4166400] first4: -0.001557 -0.000854 -0.000963 -0.000449 [Store] Unload VAE-Dec (161.1 MB) [MP3] Encoding 86.8s @ 128 kbps, 48000 Hz stereo [MP3] 1388928 bytes (12.0:1), 448 ms (193.55x realtime), 32 threads [MP3] Wrote ggml-xl-sft/request00.mp3 [Ace-Synth] All done 2026-07-12 10:07:44.798 | INFO | acestep.core.generation.handler.init_service_loader:_load_main_model_from_checkpoint:174 - [initialize_service] Attempting to load model with attention implementation: sdpa [Python] Generating (acestep-v15-xl-turbo, 8 steps)... Using precomputed LM hints Using precomputed LM hints [Python] Wrote python-xl-turbo/output.wav: 4166400 samples (86.80s @ 48kHz stereo) [Python] Done, 40 dump files [Xl-turbo] Cosine similarities GGML vs Python stage GGML vs Python text_hidden 0.999810 lyric_embed 1.000000 enc_hidden 0.998858 detok_output 0.999843 context 0.999900 noise 1.000000 temb_t 0.999905 hidden_after_proj_in 0.999931 enc_after_cond_emb 0.998778 layer0_sa_output 0.998917 hidden_after_layer0 0.999383 hidden_after_layer6 0.999592 hidden_after_layer12 0.994932 hidden_after_layer18 0.995408 hidden_after_layer31 0.987959 dit_step0_vt 0.912834 dit_step0_xt 0.999843 dit_step1_vt 0.922522 dit_step1_xt 0.999511 dit_step2_vt 0.920823 dit_step2_xt 0.998661 dit_step3_vt 0.917544 dit_step3_xt 0.996614 dit_step4_vt 0.909810 dit_step4_xt 0.991498 dit_step5_vt 0.902412 dit_step5_xt 0.979479 dit_step6_vt 0.893934 dit_step6_xt 0.955696 dit_step7_vt 0.873013 dit_x0 0.917752 vae_audio 0.702513 vae_audio (STFT cosine) 0.918481 [Xl-turbo] Error growth GGML vs Python stage cos max_err mean_err mean_A std_A mean_B std_B dit_step0_xt 0.999843 0.229904 0.012558 -0.002935 0.978421 -0.002721 0.976534 dit_step1_xt 0.999511 0.278223 0.021895 -0.005847 0.952208 -0.005567 0.949004 dit_step2_xt 0.998661 0.438987 0.034828 -0.009627 0.923853 -0.009330 0.918931 dit_step3_xt 0.996614 0.791799 0.052649 -0.015163 0.893662 -0.014418 0.887241 dit_step4_xt 0.991498 1.373815 0.079860 -0.022720 0.866009 -0.021453 0.857796 dit_step5_xt 0.979479 2.041897 0.121251 -0.033939 0.850773 -0.032240 0.840033 dit_step6_xt 0.955696 2.874097 0.184122 -0.051019 0.877586 -0.049194 0.863964 [Xl-sft] steps=50, shift=1.0 | acestep-v15-xl-sft-Q5_K_M.gguf [GGML] Running acestep-v15-xl-sft-Q5_K_M.gguf... [GGML] Done, 131 dump files [Python] Initializing acestep-v15-xl-sft... Loading checkpoint shards: 0%| | 0/4 [00:00 2026-07-12 10:07:47.888 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:123 - ====================================================================== 2026-07-12 10:07:47.888 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:124 - lyrics_text: # Languages fr # Lyric # Lyric [Intro - Guitar Riff] [Verse 1] Dans le monde des tutos virtuels G ta toise en nouvelle passion Avec Ggendoline et PumbΓ© Γ  midi La communautΓ©, c'est l'unitΓ© Quel joie, une clΓ© [Chorus] Dans le monde des tutos virtuels GΓ’ndoline et PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner [Guitar Solo] [Verse 2] Dans le monde des tutos virtuels GΓ’ndoline, PumbΓ© Γ  midi Une famille Γ  connecter, c'est vrai D'un enfant qui voit toi fusionner<|endoftext|> 2026-07-12 10:07:47.888 | INFO | acestep.core.generation.handler.conditioning_text:_prepare_text_conditioning_inputs:125 - ====================================================================== 2026-07-12 10:07:47.892 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:110 - [preprocess_batch] Inferring prompt embeddings... 2026-07-12 10:07:47.900 | INFO | acestep.core.generation.handler.conditioning_embed:preprocess_batch:113 - [preprocess_batch] Inferring lyric embeddings... 2026-07-12 10:07:47.900 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:161 - [service_generate] Generating audio... (DiT backend: PyTorch (cuda)) 2026-07-12 10:07:47.907 | INFO | acestep.core.generation.handler.service_generate_execute:_execute_service_generate_diffusion:263 - [service_generate] DiT diffusion via PyTorch (cuda)... [Python] Generating (acestep-v15-xl-sft, 50 steps)... Using precomputed LM hints Using precomputed LM hints 0%| | 0/50 [00:00