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

20
examples/passkey/passkey.cpp
View file

@ -126,6 +126,8 @@ int main(int argc, char ** argv) {
int n_past = 0;
auto * mem = llama_get_memory(ctx);
// fill the KV cache
for (int i = 0; i < n_ctx; i += n_batch) {
if (i > 0 && n_grp > 1) {
@ -133,10 +135,10 @@ int main(int argc, char ** argv) {
const int ib = i/n_batch - 1;
const int bd = n_batch_grp*(n_grp - 1);
llama_kv_self_seq_add(ctx, 0, n_past - n_batch, n_past, ib*bd);
llama_kv_self_seq_div(ctx, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
llama_memory_seq_add(mem, 0, n_past - n_batch, n_past, ib*bd);
llama_memory_seq_div(mem, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
n_past = llama_memory_seq_pos_max(mem, 0) + 1;
}
common_batch_clear(batch);
@ -166,10 +168,10 @@ int main(int argc, char ** argv) {
LOG_INF("%s: shifting KV cache with %d\n", __func__, n_discard);
llama_kv_self_seq_rm (ctx, 0, n_keep , n_keep + n_discard);
llama_kv_self_seq_add(ctx, 0, n_keep + n_discard, n_ctx, -n_discard);
llama_memory_seq_rm (mem, 0, n_keep , n_keep + n_discard);
llama_memory_seq_add(mem, 0, n_keep + n_discard, n_ctx, -n_discard);
n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
n_past = llama_memory_seq_pos_max(mem, 0) + 1;
common_batch_clear(batch);
@ -195,10 +197,10 @@ int main(int argc, char ** argv) {
if (n_discard > 0) {
LOG_INF("%s: shifting KV cache with %d to free space for the answer\n", __func__, n_discard);
llama_kv_self_seq_rm (ctx, 0, n_keep , n_keep + n_discard);
llama_kv_self_seq_add(ctx, 0, n_keep + n_discard, n_ctx, -n_discard);
llama_memory_seq_rm (mem, 0, n_keep , n_keep + n_discard);
llama_memory_seq_add(mem, 0, n_keep + n_discard, n_ctx, -n_discard);
n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
n_past = llama_memory_seq_pos_max(mem, 0) + 1;
}
}