Chirag Jain
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 mesi fa | |
|---|---|---|
| .. | ||
| README.md | 2 anni fa | |
| fused_dense.cpp | 1 anno fa | |
| fused_dense_cuda.cu | 1 anno fa | |
| setup.py | 9 mesi fa | |
README.md
This CUDA extension implements fused matmul + bias (forward and backward), and fused matmul + bias + gelu (forward and backward), adapted from Apex's FusedDense. We make it work for bfloat16.
For best performance, you should use CUDA >= 11.8. CuBLAS versions before this doesn't have the best matmul + bias + gelu performance for bfloat16.
It has only been tested on A100s.
cd csrc/fused_dense_lib && pip install .