Language Model error with vocabulary size less than 501 -> torch/backends/cudnn

Part 2 & Alumni (2018)
1

Hi Everyone,

I’m facing an error I couldn’t fix myself. When I try to run data through a Language model built with md.get_model and vocabulary size is less than 501, torch raises an error:

torch/backends/cudnn/init.py(112)init()
-> check_error(lib.cudnnCreate(ctypes.byref(ptr)))

torch/backends/cudnn/init.py in check_error(status)
281 def check_error(status):
282 if status is not 0:
–> 283 raise CuDNNError(status)

When I check what value lib.cudnnCreate(ctypes.byref(ptr)) produces I get 4.
When I change the vocabulary size to 501 or higher, this error is not raised.

Any ideas how to solve this error? Many thanks for replying!

BR, Kimmo