tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 12:05:03 +00:00
Code Issues Packages Projects Releases 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
34 changed files with 206 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/parallel/parallel.cpp
View File

@ -194,6 +194,8 @@ int main(int argc, char ** argv) {
llama_model * model = llama_init.model.get();
llama_context * ctx = llama_init.context.get();
auto * mem = llama_get_memory(ctx);
const llama_vocab * vocab = llama_model_get_vocab(model);
// load the prompts from an external file if there are any
@ -259,7 +261,7 @@ int main(int argc, char ** argv) {
// assign the system KV cache to all parallel sequences
for (int32_t i = 1; i <= n_clients; ++i) {
llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
llama_memory_seq_cp(mem, 0, i, -1, -1);
}
LOG_INF("\n");
@ -286,9 +288,9 @@ int main(int argc, char ** argv) {
if (batch.n_tokens == 0) {
// all sequences have ended - clear the entire KV cache
for (int i = 1; i <= n_clients; ++i) {
llama_kv_self_seq_rm(ctx, i, -1, -1);
llama_memory_seq_rm(mem, i, -1, -1);
// but keep the system prompt
llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
llama_memory_seq_cp(mem, 0, i, -1, -1);
}
LOG_INF("%s: clearing the KV cache\n", __func__);
@ -447,8 +449,8 @@ int main(int argc, char ** argv) {
}
// delete only the generated part of the sequence, i.e. keep the system prompt in the cache
llama_kv_self_seq_rm(ctx, client.id + 1, -1, -1);
llama_kv_self_seq_cp(ctx, 0, client.id + 1, -1, -1);
llama_memory_seq_rm(mem, client.id + 1, -1, -1);
llama_memory_seq_cp(mem, 0, client.id + 1, -1, -1);
const auto t_main_end = ggml_time_us();