Skip to content

Commit

Permalink
black
Browse files Browse the repository at this point in the history
  • Loading branch information
Andrei Panferov committed Feb 7, 2024
1 parent d90c43b commit e06a789
Showing 1 changed file with 3 additions and 1 deletion.
4 changes: 3 additions & 1 deletion inference_lib/src/aqlm/inference_kernels/numba_kernel.py
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,9 @@ def numba_gemm_lut(
assert in_features % in_group_size == 0
assert codebook_size == 2**8
assert codes.dtype == torch.int8
assert input.dtype == torch.float32 and codebooks.dtype == torch.float32, f"please load the model with `torch_dtype=torch.float32`, as {input.dtype} is not supported for CPU"
assert (
input.dtype == torch.float32 and codebooks.dtype == torch.float32
), f"please load the model with `torch_dtype=torch.float32`, as {input.dtype} is not supported for CPU"

kernel_key = (in_group_size, out_features, in_features, num_codebooks)
if kernel_key not in COMPILED_KERNELS:
Expand Down

0 comments on commit e06a789

Please sign in to comment.