tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-19 09:08:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

kv-cache : remove llama_kv_cache_i

Browse Source
This commit is contained in:
Georgi Gerganov
2025-02-19 14:36:27 +02:00
parent f5cedbcaaa
commit 5f11a5502a
5 changed files with 330 additions and 339 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/llama-graph.h
View File

@ -113,6 +113,15 @@ public:
ggml_tensor * kq,
float kq_scale) = 0;
virtual void build_kv_self_shift(
ggml_context * ctx0,
ggml_cgraph * gf) = 0;
// find holes from the beginning of the KV cache and fill them by moving data from the end of the cache
virtual void build_kv_self_defrag(
ggml_context * ctx0,
ggml_cgraph * gf) = 0;
virtual ggml_tensor * build_inp_k_shift(
ggml_context * ctx0) = 0;
@ -182,18 +191,3 @@ public:
int il,
bool worst_case) = 0;
};
class llama_graph_kv_cache_i {
public:
virtual void build_shift(
ggml_context * ctx0,
ggml_cgraph * gf,
llama_graph_i * lgf) = 0;
// find holes from the beginning of the KV cache and fill them by moving data from the end of the cache
virtual void build_defrag(
ggml_context * ctx0,
ggml_cgraph * gf,
int32_t max_nodes,
bool v_trans) = 0;
};