File size: 11,739 Bytes
46c0d05 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 |
epoch 1/20
Traceback (most recent call last):
File "/workspace/kohya-trainer/sdxl_train.py", line 649, in <module>
train(args)
File "/workspace/kohya-trainer/sdxl_train.py", line 475, in train
accelerator.backward(loss)
File "/workspace/venv/lib/python3.10/site-packages/accelerate/accelerator.py", line 1743, in backward
self.scaler.scale(loss).backward(**kwargs)
File "/workspace/venv/lib/python3.10/site-packages/torch/_tensor.py", line 487, in backward
torch.autograd.backward(
File "/workspace/venv/lib/python3.10/site-packages/torch/autograd/__init__.py", line 200, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
File "/workspace/venv/lib/python3.10/site-packages/torch/autograd/function.py", line 274, in apply
return user_fn(self, *args)
File "/workspace/venv/lib/python3.10/site-packages/torch/utils/checkpoint.py", line 157, in backward
torch.autograd.backward(outputs_with_grad, args_with_grad)
File "/workspace/venv/lib/python3.10/site-packages/torch/autograd/__init__.py", line 200, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.00 MiB (GPU 0; 23.65 GiB total capacity; 18.73 GiB already allocated; 25.69 MiB free; 19.47 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
[31m╭───────────────────── [39m[1mTraceback (most recent call last)[31m[22m ──────────────────────╮
[31m│[39m /workspace/kohya-trainer/[1msdxl_train.py[22m:[94m649[39m in [92m<module>[39m [31m│
[31m│[39m [31m│
[31m│[39m 646 │ args = parser.parse_args() [31m│
[31m│[39m 647 │ args = train_util.read_config_from_file(args, parser) [31m│
[31m│[39m 648 │ [31m│
[31m│[39m [31m❱ [39m649 │ train(args) [31m│
[31m│[39m 650 [31m│
[31m│[39m [31m│
[31m│[39m /workspace/kohya-trainer/[1msdxl_train.py[22m:[94m475[39m in [92mtrain[39m [31m│
[31m│[39m [31m│
[31m│[39m 472 │ │ │ │ [94melse[39m: [31m│
[31m│[39m 473 │ │ │ │ │ loss = torch.nn.functional.mse_loss(noise_pred.flo [31m│
[31m│[39m 474 │ │ │ │ [31m│
[31m│[39m [31m❱ [39m475 │ │ │ │ accelerator.backward(loss) [31m│
[31m│[39m 476 │ │ │ │ [94mif[39m accelerator.sync_gradients [95mand[39m args.max_grad_norm ! [31m│
[31m│[39m 477 │ │ │ │ │ params_to_clip = [] [31m│
[31m│[39m 478 │ │ │ │ │ [94mfor[39m m [95min[39m training_models: [31m│
[31m│[39m [31m│
[31m│[39m /workspace/venv/lib/python3.10/site-packages/accelerate/[1maccelerator.py[22m:[94m1743[39m [31m│
[31m│[39m in [92mbackward[39m [31m│
[31m│[39m [31m│
[31m│[39m 1740 │ │ [94melif[39m [96mself[39m.distributed_type == DistributedType.MEGATRON_LM: [31m│
[31m│[39m 1741 │ │ │ [94mreturn[39m [31m│
[31m│[39m 1742 │ │ [94melif[39m [96mself[39m.scaler [95mis[39m [95mnot[39m [94mNone[39m: [31m│
[31m│[39m [31m❱ [39m1743 │ │ │ [96mself[39m.scaler.scale(loss).backward(**kwargs) [31m│
[31m│[39m 1744 │ │ [94melse[39m: [31m│
[31m│[39m 1745 │ │ │ loss.backward(**kwargs) [31m│
[31m│[39m 1746 [31m│
[31m│[39m [31m│
[31m│[39m /workspace/venv/lib/python3.10/site-packages/torch/[1m_tensor.py[22m:[94m487[39m in [31m│
[31m│[39m [92mbackward[39m [31m│
[31m│[39m [31m│
[31m│[39m 484 │ │ │ │ create_graph=create_graph, [31m│
[31m│[39m 485 │ │ │ │ inputs=inputs, [31m│
[31m│[39m 486 │ │ │ ) [31m│
[31m│[39m [31m❱ [39m 487 │ │ torch.autograd.backward( [31m│
[31m│[39m 488 │ │ │ [96mself[39m, gradient, retain_graph, create_graph, inputs=inputs [31m│
[31m│[39m 489 │ │ ) [31m│
[31m│[39m 490 [31m│
[31m│[39m [31m│
[31m│[39m /workspace/venv/lib/python3.10/site-packages/torch/autograd/[1m__init__.py[22m:[94m200[39m [31m│
[31m│[39m in [92mbackward[39m [31m│
[31m│[39m [31m│
[31m│[39m 197 │ # The reason we repeat same the comment below is that [31m│
[31m│[39m 198 │ # some Python versions print out the first line of a multi-line fu [31m│
[31m│[39m 199 │ # calls in the traceback and some print out the last line [31m│
[31m│[39m [31m❱ [39m200 │ Variable._execution_engine.run_backward( # Calls into the C++ eng [31m│
[31m│[39m 201 │ │ tensors, grad_tensors_, retain_graph, create_graph, inputs, [31m│
[31m│[39m 202 │ │ allow_unreachable=[94mTrue[39m, accumulate_grad=[94mTrue[39m) # Calls into th [31m│
[31m│[39m 203 [31m│
[31m│[39m [31m│
[31m│[39m /workspace/venv/lib/python3.10/site-packages/torch/autograd/[1mfunction.py[22m:[94m274[39m [31m│
[31m│[39m in [92mapply[39m [31m│
[31m│[39m [31m│
[31m│[39m 271 │ │ │ │ │ │ │ [33m"Function is not allowed. You should on[39m [31m│
[31m│[39m 272 │ │ │ │ │ │ │ [33m"of them."[39m) [31m│
[31m│[39m 273 │ │ user_fn = vjp_fn [94mif[39m vjp_fn [95mis[39m [95mnot[39m Function.vjp [94melse[39m backward_f [31m│
[31m│[39m [31m❱ [39m274 │ │ [94mreturn[39m user_fn([96mself[39m, *args) [31m│
[31m│[39m 275 │ [31m│
[31m│[39m 276 │ [94mdef[39m [92mapply_jvp[39m([96mself[39m, *args): [31m│
[31m│[39m 277 │ │ # _forward_cls is defined by derived class [31m│
[31m│[39m [31m│
[31m│[39m /workspace/venv/lib/python3.10/site-packages/torch/utils/[1mcheckpoint.py[22m:[94m157[39m [31m│
[31m│[39m in [92mbackward[39m [31m│
[31m│[39m [31m│
[31m│[39m 154 │ │ │ [94mraise[39m [96mRuntimeError[39m( [31m│
[31m│[39m 155 │ │ │ │ [33m"none of output has requires_grad=True,"[39m [31m│
[31m│[39m 156 │ │ │ │ [33m" this checkpoint() is not necessary"[39m) [31m│
[31m│[39m [31m❱ [39m157 │ │ torch.autograd.backward(outputs_with_grad, args_with_grad) [31m│
[31m│[39m 158 │ │ grads = [96mtuple[39m(inp.grad [94mif[39m [96misinstance[39m(inp, torch.Tensor) [94melse[39m [94mN[39m [31m│
[31m│[39m 159 │ │ │ │ │ [94mfor[39m inp [95min[39m detached_inputs) [31m│
[31m│[39m 160 [31m│
[31m│[39m [31m│
[31m│[39m /workspace/venv/lib/python3.10/site-packages/torch/autograd/[1m__init__.py[22m:[94m200[39m [31m│
[31m│[39m in [92mbackward[39m [31m│
[31m│[39m [31m│
[31m│[39m 197 │ # The reason we repeat same the comment below is that [31m│
[31m│[39m 198 │ # some Python versions print out the first line of a multi-line fu [31m│
[31m│[39m 199 │ # calls in the traceback and some print out the last line [31m│
[31m│[39m [31m❱ [39m200 │ Variable._execution_engine.run_backward( # Calls into the C++ eng [31m│
[31m│[39m 201 │ │ tensors, grad_tensors_, retain_graph, create_graph, inputs, [31m│
[31m│[39m 202 │ │ allow_unreachable=[94mTrue[39m, accumulate_grad=[94mTrue[39m) # Calls into th [31m│
[31m│[39m 203 [31m│
[31m╰──────────────────────────────────────────────────────────────────────────────╯
[1mOutOfMemoryError: [22mCUDA out of memory. Tried to allocate [1m26.00[22m MiB [1m([22mGPU [1m0[22m; [1m23.65
GiB total capacity; [1m18.73[22m GiB already allocated; [1m25.69[22m MiB free; [1m19.47[22m GiB
reserved in total by PyTorch[1m)[22m If reserved memory is >> allocated memory try
setting max_split_size_mb to avoid fragmentation. See documentation for Memory
Management and PYTORCH_CUDA_ALLOC_CONF |