We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
KeyError: 'sum'
use_fp8=True
Expected both dimensions of mat2 to be divisible by 16 but got torch.Size([768, 2])
offload_optim_frac=1.0
min_chunk_size_m
RuntimeError: expected input to be on cuda during training
shape mismatch error: shape [32, 512] is invalid for input of size 512
AssertionError: div_scale should remain default