EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

kv-cache : support layer reuse (#15504)

Browse Source 
* kv-cache : support layer reuse

ggml-ci

* cont : update comments [no ci]
This commit is contained in:
Georgi Gerganov
2025-08-24 13:07:07 +03:00
committed by GitHub
parent c9a24fb932
commit b730706a49
12 changed files with 203 additions and 136 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/llama-hparams.h
View File

@@ -41,6 +41,7 @@ struct llama_hparams {
uint32_t n_embd;
uint32_t n_embd_features = 0;
uint32_t n_layer;
int32_t n_layer_kv_from_start = -1; // if non-negative, the first n_layer_kv_from_start layers have KV cache
uint32_t n_rot;
uint32_t n_embd_head_k; // dimension of keys (d_k). d_q is assumed to be the same, but there are n_head q heads, and only n_head_kv k-v heads
uint32_t n_embd_head_v; // dimension of values (d_v) aka n_embd_head
@@ -221,6 +222,11 @@ struct llama_hparams {
uint32_t n_pos_per_embd() const;
bool is_swa(uint32_t il) const;
bool has_kv(uint32_t il) const;
// number of layers for which has_kv() returns true
uint32_t n_layer_kv() const;
};
static_assert(std::is_trivially_copyable<llama_hparams>::value, "llama_hparams must be trivially copyable");