cuda : update supports_op for matrix multiplication #8245
ggerganov
merged
1 commit into
master
from
sl/fix-cuda-supports
cuda : update supports_op for matrix multiplication
b1b3b00f
ggerganov
approved these changes
on 2024-07-02
Login to write a write a comment.
Update
supports_opto correctly reflect that bf16 is not supported, and prevent new types added in the future from being incorrectly reported as supported. This will also cause bf16 models to be run on the CPU rather than crashing when using a CUDA build.