Setting PYTORCH_CUDA_ALLOC_CONF in backend:native,garbage_collection_threshold=0.8 #77
Open
Description
opened on Aug 26, 2024
Somehow setting below might be causing an issue to create an error from
os.environ["PYTORCH_CUDA_ALLOC_CONF"]="backend:native,garbage_collection_threshold=0.8" is causing below errors:
Traceback (most recent call last):
File "/usr/local/lib/python3.10/dist-packages/crossfit/data/dataframe/core.py", line 157, in cast
frame = CrossFrame(self.to_dict()).apply(cf.convert_array, columns)
File "/usr/local/lib/python3.10/dist-packages/crossfit/data/dataframe/core.py", line 440, in apply
data = {k: func(v, *args, **kwargs) for k, v in self.data.items()}
File "/usr/local/lib/python3.10/dist-packages/crossfit/data/dataframe/core.py", line 440, in <dictcomp>
data = {k: func(v, *args, **kwargs) for k, v in self.data.items()}
File "/usr/local/lib/python3.10/dist-packages/crossfit/data/array/conversion.py", line 110, in __call__
raise TypeError(
TypeError: Can't create 0 [1, 22868, 17081, 280, 268, 9384, 300, 341, 19...
1 [1, 879, 1071, 581, 3466, 771, 273, 389, 1351,...
2 [1, 2295, 2344, 286, 489, 4221, 321, 17728, 26...
3 [1, 5953, 289, 21333, 302, 92608, 290, 1330, 2...
4 [1, 12686, 10060, 13595, 834, 4820, 268, 263, ...
...
115819 [1, 344, 266, 1404, 600, 273, 357, 272, 4019, ...
115820 [1, 369, 760, 433, 266, 536, 5497, 265, 580, 2...
115821 [1, 279, 16762, 2517, 7677, 265, 50876, 11860,...
115822 [1, 1388, 359, 737, 8899, 293, 2967, 263, 8928...
115823 [1, 9510, 277, 3513, 559, 359, 986, 1087, 277,...
Name: input_ids, Length: 115824, dtype: list array from type <class 'torch.Tensor'>, which doesn't support any of the available conversion interfaces.
Metadata
Assignees
Labels
No labels
Activity