2026-08-29 19:54:05.474 | WARNING | acestep.training.trainer::40 - bitsandbytes not installed. Using standard AdamW. ░▒▓███████▓▒░▒▓█▓▒░▒▓███████▓▒░░▒▓████████▓▒░░▒▓███████▓▒░▒▓████████▓▒░▒▓████████▓▒░▒▓███████▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░░▒▓█▓▒░ ░▒▓██████▓▒░░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓██████▓▒░ ░▒▓██████▓▒░ ░▒▓█▓▒░ ░▒▓██████▓▒░ ░▒▓███████▓▒░ ░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓███████▓▒░░▒▓█▓▒░▒▓███████▓▒░░▒▓████████▓▒░▒▓███████▓▒░ ░▒▓█▓▒░ ░▒▓████████▓▒░▒▓█▓▒░ "Research grade? No. dernet grade." Side-Step v2.0.0 -- Adapter Fine-Tuning CLI (LoRA + LoKR) Standalone: github.com/koda-dernet/Side-Step Mode : fixed (corrected timesteps + CFG dropout) Stack : Python 3.12.14 | PyTorch 2.10.0+cu128 | CUDA 12.8 | bf16 GPU : NVIDIA GeForce RTX 3090 (23.6 GiB) ============================================================ Training Configuration ============================================================ [Model] Model variant........... turbo Checkpoint dir.......... music/checkpoints * Dataset dir............. music/tensors/pd-shakedown/20260829T162506Z-stage0 [Device] Device.................. cuda:0 * Precision............... bf16 * [LoRA] Rank (r)................ 64 Alpha................... 128 Dropout................. 0.1 Target modules.......... q_proj, k_proj, v_proj, o_proj Bias.................... none [Training] Learning rate........... 1.0e-04 Batch size.............. 1 Grad accumulation....... 4 Effective batch......... 4 Max epochs.............. 10 * Warmup steps............ 100 Weight decay............ 0.01 Max grad norm........... 1 Seed.................... 42 [Corrected Training] CFG dropout ratio....... 0.15 Timestep mu............. -0.4 Timestep sigma.......... 1 Data proportion......... 0.5 [Checkpointing] Output dir.............. music/adapters/pd-shakedown/20260829T162506Z-stage0-r64 * Save every N epochs..... 5 * Resume from............. (auto) [Logging] TensorBoard dir......... (auto) Log every N steps....... 10 Grad norms every N steps 50 ============================================================ (* = non-default value) ============================================================ [INFO] Loading model (variant=turbo, device=cuda:0) 2026-08-29 19:54:05 [INFO] acestep.training_v2.model_loader: [INFO] Loading model from music/checkpoints/acestep-v15-turbo (variant=turbo, dtype=torch.bfloat16) 2026-08-29 19:54:07 [INFO] acestep.training_v2.model_loader: [OK] Model on cuda:0 (torch.bfloat16), all params frozen 2026-08-29 19:54:08.184 | INFO | acestep.training.lora_injection:inject_lora_into_dit:232 - LoRA injected into DiT decoder: 2026-08-29 19:54:08.184 | INFO | acestep.training.lora_injection:inject_lora_into_dit:233 - Total parameters: 2,437,912,710 2026-08-29 19:54:08.184 | INFO | acestep.training.lora_injection:inject_lora_into_dit:234 - Trainable parameters: 44,040,192 (1.81%) 2026-08-29 19:54:08.184 | INFO | acestep.training.lora_injection:inject_lora_into_dit:237 - LoRA rank: 64, alpha: 128 2026-08-29 19:54:08 [INFO] acestep.training_v2.fixed_lora_module: [OK] LoRA injected: 44,040,192 trainable params 2026-08-29 19:54:08.185 | INFO | acestep.training.data_module:__init__:90 - PreprocessedTensorDataset: 19 samples from music/tensors/pd-shakedown/20260829T162506Z-stage0 INFO: Using bfloat16 Automatic Mixed Precision (AMP) 2026-08-29 19:54:08 [INFO] lightning.pytorch.utilities.rank_zero: Using bfloat16 Automatic Mixed Precision (AMP) 2026-08-29 19:54:08 [INFO] acestep.training_v2.tensorboard_utils: [OK] TensorBoard logger initialised at music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/runs 2026-08-29 19:54:08 [INFO] acestep.training_v2.optim: [Side-Step] Using AdamW optimizer 2026-08-29 19:54:08.199 | INFO | acestep.training.lora_injection:_safe_enable_input_require_grads:50 - Skipping enable_input_require_grads for decoder: get_input_embeddings is not implemented (expected for DiT) [INFO] Loading model from music/checkpoints/acestep-v15-turbo (variant=turbo, dtype=torch.bfloat16) [OK] Model loaded with attn_implementation=sdpa [OK] Loaded 19 preprocessed samples [INFO] Starting training (devices: 1, strategy: auto, precision: bf16-mixed) [INFO] Training 44,040,192 parameters [INFO] Optimizer: adamw [INFO] Scheduler: cosine [INFO] Gradient checkpointing enabled (use_cache=False, input_grads=False) 2026-08-29 19:56:32 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/checkpoints/epoch_5_loss_0.6587 2026-08-29 19:56:32 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/checkpoints/epoch_5_loss_0.6587 (epoch 5, step 25) 2026-08-29 19:58:56 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/checkpoints/epoch_10_loss_0.6101 2026-08-29 19:58:56 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/checkpoints/epoch_10_loss_0.6101 (epoch 10, step 50) 2026-08-29 19:58:56 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/final 2026-08-29 19:58:56 [INFO] acestep.training_v2.trainer_helpers: [OK] Adapter verified: 44,040,192 params, 44,040,192 non-zero (100.0%), max|w|=0.024902 ============================================================ Training Complete ============================================================ Total time .......... 4m 49s Epochs .............. 0 / 10 Total steps ......... 50 Loss ................ 0.9383 -> 0.6372 Best loss ........... 0.5666 Peak VRAM ........... 5.7 GiB Saved Checkpoints: Epoch 5 Loss: 0.6587 Epoch 10 Loss: 0.6101 Output dir .......... music/adapters/pd-shakedown/20260829T162506Z-stage0-r64 Final weights ....... music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/final (84.1 MiB) TensorBoard ......... music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/runs ============================================================ [OK] Epoch 1/10 in 29.0s, Loss: 0.9383 Epoch 2/10, Step 10, Loss: 0.8265 [OK] Epoch 2/10 in 28.4s, Loss: 0.8123 [OK] Epoch 3/10 in 28.6s, Loss: 0.6856 Epoch 4/10, Step 20, Loss: 0.6084 [OK] Epoch 4/10 in 28.7s, Loss: 0.6633 [OK] Epoch 5/10 in 28.8s, Loss: 0.6587 [OK] Checkpoint saved at epoch 5 Epoch 6/10, Step 30, Loss: 0.6160 [OK] Epoch 6/10 in 28.8s, Loss: 0.6476 [OK] Epoch 7/10 in 28.8s, Loss: 0.6424 Epoch 8/10, Step 40, Loss: 0.5666 [OK] Epoch 8/10 in 28.8s, Loss: 0.6295 [OK] Epoch 9/10 in 28.8s, Loss: 0.6598 Epoch 10/10, Step 50, Loss: 0.6135 [OK] Epoch 10/10 in 28.8s, Loss: 0.6101 [OK] Checkpoint saved at epoch 10 [OK] Training complete! LoRA saved to music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/final For inference, set your LoRA path to: music/adapters/pd-shakedown/20260829T162506Z-stage0-r64/final