tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-19 17:17:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

graph : simplify attention api

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-02-19 18:43:49 +02:00
parent e17e4b72d1
commit 2eacb4c1bf
4 changed files with 47 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
src/llama-context.h
View File

@ -376,20 +376,13 @@ public:
bool swa,
bool worst_case) override;
virtual void build_attn_kv_store(
ggml_context * ctx0,
ggml_cgraph * gf,
ggml_tensor * k_cur,
ggml_tensor * v_cur,
int32_t n_tokens,
int64_t il,
bool worst_case) override;
virtual ggml_tensor * build_attn_qkv(
virtual ggml_tensor * build_attn(
ggml_context * ctx0,
ggml_cgraph * gf,
ggml_tensor * wo,
ggml_tensor * wo_b,
ggml_tensor * k_cur,
ggml_tensor * v_cur,
ggml_tensor * q_cur,
int32_t n_tokens,
float kq_scale,
@ -443,6 +436,7 @@ protected:
// a recurrent transformer (ie.e RWKV, Mamba)
// TODO: temporary reuse kv_self, but in the future, implement recurrent-specific context with specific cache
//class llama_context_recurrent : public llama_context {
class llama_context_recurrent : public llama_context_kv_self {
public:
llama_context_recurrent(