Skip to content

CUDA: fix RoPE asserts, block sizes #2833

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 28, 2023
Merged

CUDA: fix RoPE asserts, block sizes #2833

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions ggml-cuda.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4908,16 +4908,17 @@ static void scale_f32_cuda(const float * x, float * dst, const float scale, cons

static void rope_f32_cuda(const float * x, float * dst, const int ncols, const int nrows, const float p0,
const float p_delta, const int p_delta_rows, const float theta_scale, cudaStream_t stream) {
GGML_ASSERT(nrows % 2 == 0); // GG: is this assert really needed? I don't see why
const dim3 block_dims(1, 2*CUDA_ROPE_BLOCK_SIZE, 1);
GGML_ASSERT(ncols % 2 == 0);
const dim3 block_dims(1, CUDA_ROPE_BLOCK_SIZE, 1);
const int num_blocks_x = (ncols + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE);
const dim3 block_nums(nrows, num_blocks_x, 1);
rope_f32<<<block_nums, block_dims, 0, stream>>>(x, dst, ncols, p0, p_delta, p_delta_rows, theta_scale);
}

static void rope_neox_f32_cuda(const float * x, float * dst, const int ncols, const int nrows, const float p0,
const float p_delta, const int p_delta_rows, const float theta_scale, cudaStream_t stream) {
const dim3 block_dims(1, 2*CUDA_ROPE_BLOCK_SIZE, 1);
GGML_ASSERT(ncols % 2 == 0);
const dim3 block_dims(1, CUDA_ROPE_BLOCK_SIZE, 1);
const int num_blocks_x = (ncols + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE);
const dim3 block_nums(nrows, num_blocks_x, 1);
rope_neox_f32<<<block_nums, block_dims, 0, stream>>>(x, dst, ncols, p0, p_delta, p_delta_rows, theta_scale);
Expand Down