Skip to content
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

llama : pad KV cache size #4280

Merged
merged 2 commits into from
Dec 3, 2023
Merged

llama : pad KV cache size #4280

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3e68df8
llama : pad KV cache size to 32
ggerganov Dec 1, 2023
3cb1c34
metal : try to improve batched decoding
ggerganov Dec 1, 2023
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
2 changes: 1 addition & 1 deletion ggml-metal.m
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1083,7 +1083,7 @@ void ggml_metal_graph_compute(

// find the break-even point where the matrix-matrix kernel becomes more efficient compared
// to the matrix-vector kernel
int ne11_mm_min = 1;
int ne11_mm_min = src0t == GGML_TYPE_F16 ? 1 : 16;

#if 0
// the numbers below are measured on M2 Ultra for 7B and 13B models
Expand Down
3 changes: 1 addition & 2 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5744,8 +5744,7 @@ static int llama_decode_internal(
// a heuristic, to avoid attending the full cache if it is not yet utilized
// after enough generations, the benefit from this heuristic disappears
// if we start defragmenting the cache, the benefit from this will be more important
//kv_self.n = std::max(32, GGML_PAD(llama_kv_cache_cell_max(kv_self), 32)); // TODO: this might be better for CUDA?
kv_self.n = std::min((int32_t) cparams.n_ctx, std::max(32, llama_kv_cache_cell_max(kv_self)));
kv_self.n = std::min((int32_t) cparams.n_ctx, std::max(32, GGML_PAD(llama_kv_cache_cell_max(kv_self), 32)));

//printf("kv_self.n = %5d, kv_self.used = %5d, kv_self.head = %5d\n", kv_self.n, kv_self.used, kv_self.head);

Expand Down
Loading