TypeError when using cuda about muse HOT 4 CLOSED

I am getting the same error with cuda set to false
TypeError: tensor(0.6204) is not JSON serializable

Traceback (most recent call last):
File "supervised.py", line 93, in
logger.info("log:%s" % json.dumps(to_log))
File "/usr/lib/python2.7/json/init.py", line 244, in dumps
return _default_encoder.encode(obj)
File "/usr/lib/python2.7/json/encoder.py", line 207, in encode
chunks = self.iterencode(o, _one_shot=True)
File "/usr/lib/python2.7/json/encoder.py", line 270, in iterencode
return _iterencode(o, 0)
File "/usr/lib/python2.7/json/encoder.py", line 184, in default
raise TypeError(repr(o) + " is not JSON serializable")

from muse.

Which version of PyTorch are you using?

from muse.

PyTorch version: 0.4.0
CUDA version: V9.0.176
Python version: 3.6

from muse.

This is most likely due to the last version of PyTorch -- many changes appeared in the version 0.4.0 (actually I'm quite surprised the script did not fail before line 93).

Looking at the error it looks like there is something in to_log which is a PyTorch object (maybe a scalar Tensor, as opposed to a float with PyTorch 0.3), and json does not know how to serialize this. If this is the issue then this should be a simple fix. Can you print everything in to_log and copy-paste it here? My guess is that there is a Torch tensor somewhere that should just be a Python scalar. This https://discuss.pytorch.org/t/get-value-out-of-torch-cuda-float-tensor/2539/7 might also be relevant (post is outdated, but the issue would be similar).

I don't have time to look more into this now, but we will push a fix next week.

from muse.