The assertion outputs "Cannot use AMP for PyTorch version < 1.6" however, the way the PyTorch_over_1_6 check is handled, is contrary to this, causing it to fail with Pytorch version 1.11 (Included in the conda env)
The test was indeed not clean, thx for pointing out
However, the fix you proposed was not really solving the issue
I think the best is just to comment this part; we did it on main so I can close this PR
As a side note, this part of the code was used for the old training loop; now there is a more effective way to train SPLADE models with HuggingFace trainers (in hf_train.py)
The assertion outputs
"Cannot use AMP for PyTorch version < 1.6"
however, the way thePyTorch_over_1_6
check is handled, is contrary to this, causing it to fail with Pytorch version 1.11 (Included in the conda env)