Update README.md
#2
by
kreier
- opened
Included link to the QAT - Quantization Aware Training. And the quote relating to bfloat16 is now included.
I just finished a shortened bfloat16 definition for CUDA in another project. Down to 5 lines of code from over 5000 lines (/usr/local/cuda/include/cuda_bf16.h), provided by Nvidia. And it works!
Looks good!
kreier
changed pull request status to
merged