Open
Description
Describe the bug
When using `autocast`` in dreamer example, there is a RuntimeError:
RuntimeError: masked_scatter_: expected self and source to have same dtypes but gotHalf and Float
Unfortunately, its seem to be a bug of PyTorch itself. (pytorch/pytorch#81876)
To Reproduce
Run dreamer example.
Whole output:
collector: MultiaSyncDataCollector()
init seed: 42, final seed: 971637020
7%|████████▌ | 36800/500000 [20:23<4:48:42, 26.74it/s]
Error executing job with overrides: []
Traceback (most recent call last):
File "/home/frank/Projects/rl_dev/examples/dreamer/dreamer.py", line 359, in main
scaler2.scale(actor_loss_td["loss_actor"]).backward()
File "/home/frank/anaconda3/envs/rl_dev/lib/python3.9/site-packages/torch/_tensor.py", line 492, in backward
torch.autograd.backward(
File "/home/frank/anaconda3/envs/rl_dev/lib/python3.9/site-packages/torch/autograd/__init__.py", line 251, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: masked_scatter_: expected self and source to have same dtypes but gotHalf and Float
System info
Describe the characteristic of your environment:
- Describe how the library was installed (pip, source, ...)
- Python version
- Versions of any other relevant libraries
import torchrl, numpy, sys
print(torchrl.__version__, numpy.__version__, sys.version, sys.platform)0.2.1 1.26.1 3.9.18 (main, Sep 11 2023, 13:41:44)
[GCC 11.2.0] linux
Reason and Possible fixes
Maybe we should disable autocast until this bug is fixed by torch?
Checklist
- I have checked that there is no similar issue in the repo (required)
- I have read the documentation (required)
- I have provided a minimal working example to reproduce the bug (required)