2026-09-06 09:52:06.474 | WARNING | acestep.training.trainer::40 - bitsandbytes not installed. Using standard AdamW. ░▒▓███████▓▒░▒▓█▓▒░▒▓███████▓▒░░▒▓████████▓▒░░▒▓███████▓▒░▒▓████████▓▒░▒▓████████▓▒░▒▓███████▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░░▒▓█▓▒░ ░▒▓██████▓▒░░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓██████▓▒░ ░▒▓██████▓▒░ ░▒▓█▓▒░ ░▒▓██████▓▒░ ░▒▓███████▓▒░ ░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░▒▓█▓▒░▒▓█▓▒░░▒▓█▓▒░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓█▓▒░ ░▒▓███████▓▒░░▒▓█▓▒░▒▓███████▓▒░░▒▓████████▓▒░▒▓███████▓▒░ ░▒▓█▓▒░ ░▒▓████████▓▒░▒▓█▓▒░ "Red errors are just decoration." 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/house-wide/20260906T092541Z-house0 [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.............. 5 * 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/house-wide/20260906T092541Z-house0-r64 * Save every N epochs..... 1 * 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) 2026-09-06 09:52:06 [INFO] acestep.training_v2.model_loader: [INFO] Loading model from music/checkpoints/acestep-v15-turbo (variant=turbo, dtype=torch.bfloat16) 2026-09-06 09:52:08 [INFO] acestep.training_v2.model_loader: [OK] Model on cuda:0 (torch.bfloat16), all params frozen 2026-09-06 09:52:09.333 | INFO | acestep.training.lora_injection:inject_lora_into_dit:232 - LoRA injected into DiT decoder: 2026-09-06 09:52:09.333 | INFO | acestep.training.lora_injection:inject_lora_into_dit:233 - Total parameters: 2,437,912,710 2026-09-06 09:52:09.334 | INFO | acestep.training.lora_injection:inject_lora_into_dit:234 - Trainable parameters: 44,040,192 (1.81%) 2026-09-06 09:52:09.334 | INFO | acestep.training.lora_injection:inject_lora_into_dit:237 - LoRA rank: 64, alpha: 128 2026-09-06 09:52:09 [INFO] acestep.training_v2.fixed_lora_module: [OK] LoRA injected: 44,040,192 trainable params 2026-09-06 09:52:09.346 | INFO | acestep.training.data_module:__init__:90 - PreprocessedTensorDataset: 331 samples from music/tensors/house-wide/20260906T092541Z-house0 INFO: Using bfloat16 Automatic Mixed Precision (AMP) 2026-09-06 09:52:09 [INFO] lightning.pytorch.utilities.rank_zero: Using bfloat16 Automatic Mixed Precision (AMP) 2026-09-06 09:52:09 [INFO] acestep.training_v2.tensorboard_utils: [OK] TensorBoard logger initialised at music/adapters/house-wide/20260906T092541Z-house0-r64/runs 2026-09-06 09:52:09 [INFO] acestep.training_v2.optim: [Side-Step] Using AdamW optimizer 2026-09-06 09:52:09.360 | 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 331 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) 2026-09-06 10:00:04 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_1_loss_0.8833 2026-09-06 10:00:05 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_1_loss_0.8833 (epoch 1, step 83) 2026-09-06 10:08:00 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_2_loss_0.6778 2026-09-06 10:08:01 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_2_loss_0.6778 (epoch 2, step 166) 2026-09-06 10:15:57 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_3_loss_0.6480 2026-09-06 10:15:57 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_3_loss_0.6480 (epoch 3, step 249) 2026-09-06 10:23:53 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_4_loss_0.6494 2026-09-06 10:23:53 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_4_loss_0.6494 (epoch 4, step 332) 2026-09-06 10:31:49 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_5_loss_0.6494 2026-09-06 10:31:49 [INFO] acestep.training_v2.trainer_helpers: Training checkpoint saved to music/adapters/house-wide/20260906T092541Z-house0-r64/checkpoints/epoch_5_loss_0.6494 (epoch 5, step 415) 2026-09-06 10:31:49 [INFO] acestep.training_v2.trainer_helpers: [OK] LoRA adapter saved to music/adapters/house-wide/20260906T092541Z-house0-r64/final 2026-09-06 10:31:49 [INFO] acestep.training_v2.trainer_helpers: [OK] Adapter verified: 44,040,192 params, 44,040,192 non-zero (100.0%), max|w|=0.034424 ============================================================ Training Complete ============================================================ Total time .......... 39m 40s Epochs .............. 0 / 5 Total steps ......... 415 Loss ................ 1.2842 -> 0.7743 Best loss ........... 0.4190 Peak VRAM ........... 5.7 GiB Saved Checkpoints: Epoch 1 Loss: 0.8833 Epoch 2 Loss: 0.6778 Epoch 3 Loss: 0.6480 Epoch 4 Loss: 0.6494 Epoch 5 Loss: 0.6494 Output dir .......... music/adapters/house-wide/20260906T092541Z-house0-r64 Final weights ....... music/adapters/house-wide/20260906T092541Z-house0-r64/final (84.1 MiB) TensorBoard ......... music/adapters/house-wide/20260906T092541Z-house0-r64/runs ============================================================ Epoch 1/5, Step 10, Loss: 1.2842 Epoch 1/5, Step 20, Loss: 0.9155 Epoch 1/5, Step 30, Loss: 0.9125 Epoch 1/5, Step 40, Loss: 0.7547 Epoch 1/5, Step 50, Loss: 0.7663 Epoch 1/5, Step 60, Loss: 0.7385 Epoch 1/5, Step 70, Loss: 0.6823 Epoch 1/5, Step 80, Loss: 0.6950 [OK] Epoch 1/5 in 475.2s, Loss: 0.8833 [OK] Checkpoint saved at epoch 1 Epoch 2/5, Step 90, Loss: 0.6590 Epoch 2/5, Step 100, Loss: 0.8224 Epoch 2/5, Step 110, Loss: 0.6774 Epoch 2/5, Step 120, Loss: 0.6956 Epoch 2/5, Step 130, Loss: 0.7698 Epoch 2/5, Step 140, Loss: 0.6866 Epoch 2/5, Step 150, Loss: 0.7262 Epoch 2/5, Step 160, Loss: 0.6309 [OK] Epoch 2/5 in 475.7s, Loss: 0.6778 [OK] Checkpoint saved at epoch 2 Epoch 3/5, Step 170, Loss: 0.5962 Epoch 3/5, Step 180, Loss: 0.6968 Epoch 3/5, Step 190, Loss: 0.6760 Epoch 3/5, Step 200, Loss: 0.7380 Epoch 3/5, Step 210, Loss: 0.7535 Epoch 3/5, Step 220, Loss: 0.6317 Epoch 3/5, Step 230, Loss: 0.6580 Epoch 3/5, Step 240, Loss: 0.6836 [OK] Epoch 3/5 in 475.8s, Loss: 0.6480 [OK] Checkpoint saved at epoch 3 Epoch 4/5, Step 250, Loss: 0.7379 Epoch 4/5, Step 260, Loss: 0.6962 Epoch 4/5, Step 270, Loss: 0.7221 Epoch 4/5, Step 280, Loss: 0.6117 Epoch 4/5, Step 290, Loss: 0.5079 Epoch 4/5, Step 300, Loss: 0.6238 Epoch 4/5, Step 310, Loss: 0.6299 Epoch 4/5, Step 320, Loss: 0.7127 Epoch 4/5, Step 330, Loss: 0.6108 [OK] Epoch 4/5 in 475.8s, Loss: 0.6494 [OK] Checkpoint saved at epoch 4 Epoch 5/5, Step 340, Loss: 0.6998 Epoch 5/5, Step 350, Loss: 0.8785 Epoch 5/5, Step 360, Loss: 0.6639 Epoch 5/5, Step 370, Loss: 0.6449 Epoch 5/5, Step 380, Loss: 0.7319 Epoch 5/5, Step 390, Loss: 0.7164 Epoch 5/5, Step 400, Loss: 0.4190 Epoch 5/5, Step 410, Loss: 0.7623 [OK] Epoch 5/5 in 475.7s, Loss: 0.6494 [OK] Checkpoint saved at epoch 5 [OK] Training complete! LoRA saved to music/adapters/house-wide/20260906T092541Z-house0-r64/final For inference, set your LoRA path to: music/adapters/house-wide/20260906T092541Z-house0-r64/final