ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

llama : deprecate llama_kv_self_ API (#14030)

Browse source 
* llama : deprecate llama_kv_self_ API

ggml-ci

* llama : allow llama_memory_(nullptr)

ggml-ci

* memory : add flag for optional data clear in llama_memory_clear

ggml-ci
This commit is contained in:
Georgi Gerganov 2025-06-06 14:11:15 +03:00 • committed by GitHub
parent 487a5e0401
commit 745aa5319b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
34 changed files with 206 additions and 127 deletions
Download patch file Download diff file Expand all files Collapse all files

8
examples/llama.android/llama/src/main/cpp/llama-android.cpp
View file

@ -194,7 +194,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
}
batch->logits[batch->n_tokens - 1] = true;
llama_kv_self_clear(context);
llama_memory_clear(llama_get_memory(context), false);
const auto t_pp_start = ggml_time_us();
if (llama_decode(context, *batch) != 0) {
@ -206,7 +206,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
LOGi("Benchmark text generation (tg)");
llama_kv_self_clear(context);
llama_memory_clear(llama_get_memory(context), false);
const auto t_tg_start = ggml_time_us();
for (i = 0; i < tg; i++) {
@ -223,7 +223,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
const auto t_tg_end = ggml_time_us();
llama_kv_self_clear(context);
llama_memory_clear(llama_get_memory(context), false);
const auto t_pp = double(t_pp_end - t_pp_start) / 1000000.0;
const auto t_tg = double(t_tg_end - t_tg_start) / 1000000.0;
@ -448,5 +448,5 @@ Java_android_llama_cpp_LLamaAndroid_completion_1loop(
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_kv_1cache_1clear(JNIEnv *, jobject, jlong context) {
llama_kv_self_clear(reinterpret_cast<llama_context *>(context));
llama_memory_clear(llama_get_memory(reinterpret_cast<llama_context *>(context)), true);
}