Skip to content

fix(gguf): chunk the moe_vec z grid past the 65535 cap #211

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

Open
avlp12 wants to merge 1 commit into
base: main
Choose a base branch
from
+316 −351
Open

fix(gguf): chunk the moe_vec z grid past the 65535 cap #211

Show file tree
Hide file tree
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: 6 additions & 1 deletion python/freetoken/kernel/csrc/gguf/gguf_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,12 @@ static void quantize_row_q8_1_cuda(const scalar_t* x, void* vy, const int kx, co
const dim3 num_blocks(block_num_x, num_blocks_y, 1);
const dim3 block_size(CUDA_DEQUANTIZE_BLOCK_SIZE, 1, 1);
quantize_q8_1<<<num_blocks, block_size, 0, stream>>>(
&x[off * kx], (int32_t*)vy + off * (kx_padded / 32 * 9), kx, kx_padded);
// int64: ``off`` walks up to ``ky``, which on the MoE down-projection is
// ``tokens * top_k``. ``off * kx`` wraps int32 past off = 524288 at
// kx = 4096, which would turn a launch that now succeeds into a silently
// wrong read. (``kx_padded`` is already int64, so the ``vy`` term is not
// affected.)
&x[(int64_t)off * (int64_t)kx], (int32_t*)vy + off * (kx_padded / 32 * 9), kx, kx_padded);
}
}

Expand Down
Loading