nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

ggml-cpu: FA split across kv for faster TG (#19209)

Browse Source 
* ggml-cpu: split across kv for faster TG

* simplify sinks application

* add ref impl
This commit is contained in:
Aman Gupta
2026-02-03 01:19:55 +08:00
committed by GitHub
parent a3fa035822
commit 9f682fb640
6 changed files with 220 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/include/ggml-cpu.h
View File

@@ -19,6 +19,9 @@ extern "C" {
// abort ggml_graph_compute when true
ggml_abort_callback abort_callback;
void * abort_callback_data;
// use only reference implementations
bool use_ref;
};
// numa strategies
@@ -132,6 +135,8 @@ extern "C" {
GGML_BACKEND_API void ggml_backend_cpu_set_threadpool (ggml_backend_t backend_cpu, ggml_threadpool_t threadpool);
GGML_BACKEND_API void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
GGML_BACKEND_API void ggml_backend_cpu_set_use_ref(ggml_backend_t backend_cpu, bool use_ref);
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_cpu_reg(void);
GGML_BACKEND_API void ggml_cpu_fp32_to_fp32(const float *, float *, int64_t);