-
Notifications
You must be signed in to change notification settings - Fork 31
Description
I want to combine Distrifuser with torch compile but it doesn't work
pipeline.pipeline.unet = torch.compile(pipeline.pipeline.unet)
Here is error:
[00:00<?, ?it/s][rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] Graph break from Tensor.item(), consider setting:
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] torch._dynamo.config.capture_scalar_outputs = True
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] or:
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] env TORCHDYNAMO_CAPTURE_SCALAR_OUTPUTS=1
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] to include these operations in the captured graph.
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0]
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] Graph break: from user code at:
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] File "/workspace/SD-parrallel/distrifuser/distrifuser/models/distri_sdxl_unet_pp.py", line 94, in forward
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] static_inputs["timestep"][b] = timestep.item()
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0]
[rank0]:W0202 03:40:07.939000 7752 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0]
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] Graph break from Tensor.item(), consider setting:
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] torch._dynamo.config.capture_scalar_outputs = True
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] or:
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] env TORCHDYNAMO_CAPTURE_SCALAR_OUTPUTS=1
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] to include these operations in the captured graph.
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0]
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] Graph break: from user code at:
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] File "/workspace/SD-parrallel/distrifuser/distrifuser/models/distri_sdxl_unet_pp.py", line 94, in forward
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0] static_inputs["timestep"][b] = timestep.item()
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0]
[rank1]:W0202 03:40:08.147000 7753 site-packages/torch/_dynamo/variables/tensor.py:776] [0/0]
Any idea to fix this?