ankan-ban / llama2.cu

Inference Llama 2 in one file of pure Cuda
MIT License
16 stars 2 forks source link

rename variables #3

Closed kroggen closed 1 year ago

kroggen commented 1 year ago

These changes make the code more understandable

kroggen commented 1 year ago

I added an additional commit that decrease computation by pre-computing weight[index * weight_row_stride] and putting the result in weight

kroggen commented 1 year ago

There is both mat_vec_kernel and vec_mat_kernel functions

It would be better to have more meaningful names

ankan-ban commented 1 year ago

Thank you. Certainly more readable. I will merge this soon (if you can fix the conflicts).

kroggen commented 1 year ago

The conflicts were fixed