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

13
src/llama-kv-cache-recurrent.cpp
View file

@ -117,18 +117,21 @@ llama_kv_cache_recurrent::llama_kv_cache_recurrent(
}
}
void llama_kv_cache_recurrent::clear() {
void llama_kv_cache_recurrent::clear(bool data) {
for (int32_t i = 0; i < (int32_t) size; ++i) {
cells[i].pos = -1;
cells[i].seq_id.clear();
cells[i].src = -1;
cells[i].tail = -1;
}
head = 0;
used = 0;
for (auto & buf : bufs) {
ggml_backend_buffer_clear(buf.get(), 0);
if (data) {
for (auto & buf : bufs) {
ggml_backend_buffer_clear(buf.get(), 0);
}
}
}
@ -723,7 +726,7 @@ void llama_kv_cache_recurrent::state_read(llama_io_read_i & io, llama_seq_id seq
if (!res) {
if (seq_id == -1) {
clear();
clear(true);
} else {
seq_rm(seq_id, -1, -1);
}
@ -880,7 +883,7 @@ bool llama_kv_cache_recurrent::state_read_meta(llama_io_read_i & io, uint32_t ce
return false;
}
clear();
clear(true);
for (uint32_t i = 0; i < cell_count; ++i) {
kv_cell & cell = cells[i];