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

llama : remove llama_kv_cache_view API + remove deprecated (#13653)

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-05-20 16:13:16 +03:00 • committed by GitHub
parent b69f1647f9
commit a4090d1174
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
10 changed files with 1 additions and 390 deletions
Download patch file Download diff file Expand all files Collapse all files

13
examples/lookahead/lookahead.cpp
View file

@ -50,8 +50,6 @@ int main(int argc, char ** argv) {
const int N = 5; // n-gram size
const int G = 15; // max verification n-grams
const bool dump_kv_cache = params.dump_kv_cache;
// init llama.cpp
llama_backend_init();
llama_numa_init(params.numa);
@ -152,9 +150,6 @@ int main(int argc, char ** argv) {
// here we keep adding new n-grams as we go
ngram_container ngrams_observed(llama_vocab_n_tokens(vocab), N, G);
// debug
struct llama_kv_cache_view kvc_view = llama_kv_cache_view_init(ctx, W + G + 1);
const auto t_dec_start = ggml_time_us();
// sample first token
@ -172,12 +167,6 @@ int main(int argc, char ** argv) {
}
while (true) {
// debug
if (dump_kv_cache) {
llama_kv_cache_view_update(ctx, &kvc_view);
common_kv_cache_dump_view_seqs(kvc_view, 40);
}
// build the mask from https://lmsys.org/blog/2023-11-21-lookahead-decoding/
//
// Example for W = 5, N = 4, G = 2:
@ -473,8 +462,6 @@ int main(int argc, char ** argv) {
common_sampler_free(smpl);
llama_kv_cache_view_free(&kvc_view);
llama_batch_free(batch);
llama_backend_free();

11
examples/lookup/lookup.cpp
View file

@ -24,8 +24,6 @@ int main(int argc, char ** argv){
// max. number of additional tokens to draft if match is found
const int n_draft = params.speculative.n_max;
const bool dump_kv_cache = params.dump_kv_cache;
// init llama.cpp
llama_backend_init();
llama_numa_init(params.numa);
@ -110,18 +108,9 @@ int main(int argc, char ** argv){
llama_batch batch_tgt = llama_batch_init(params.n_ctx, 0, 1);
// debug
struct llama_kv_cache_view kvc_view = llama_kv_cache_view_init(ctx, 1);
const auto t_dec_start = ggml_time_us();
while (true) {
// debug
if (dump_kv_cache) {
llama_kv_cache_view_update(ctx, &kvc_view);
common_kv_cache_dump_view_seqs(kvc_view, 40);
}
// print current draft sequence
LOG_DBG("drafted %s\n", string_from(ctx, draft).c_str());

9
examples/parallel/parallel.cpp
View file

@ -178,8 +178,6 @@ int main(int argc, char ** argv) {
// insert new requests as soon as the previous one is done
const bool cont_batching = params.cont_batching;
const bool dump_kv_cache = params.dump_kv_cache;
// is the system prompt shared in the cache
const bool is_sp_shared = params.is_pp_shared;
@ -241,8 +239,6 @@ int main(int argc, char ** argv) {
int32_t n_total_gen = 0;
int32_t n_cache_miss = 0;
struct llama_kv_cache_view kvc_view = llama_kv_cache_view_init(ctx, n_clients);
const auto t_main_start = ggml_time_us();
LOG_INF("%s: Simulating parallel requests from clients:\n", __func__);
@ -272,11 +268,6 @@ int main(int argc, char ** argv) {
LOG_INF("Processing requests ...\n\n");
while (true) {
if (dump_kv_cache) {
llama_kv_cache_view_update(ctx, &kvc_view);
common_kv_cache_dump_view_seqs(kvc_view, 40);
}
common_batch_clear(batch);
// decode any currently ongoing sequences