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

llama : add struct llama_kv_cache (wip) [no ci]

Browse source
This commit is contained in:
Georgi Gerganov 2025-01-13 14:13:11 +02:00
parent 178a7eb952
commit f78b396ee7
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
8 changed files with 428 additions and 415 deletions
Download patch file Download diff file Expand all files Collapse all files

5
examples/embedding/embedding.cpp
View file

@ -34,10 +34,11 @@ static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & toke
static void batch_decode(llama_context * ctx, llama_batch & batch, float * output, int n_seq, int n_embd, int embd_norm) {
const enum llama_pooling_type pooling_type = llama_pooling_type(ctx);
const struct llama_model * model = llama_get_model(ctx);
const llama_model * model = llama_get_model(ctx);
llama_kv_cache * kv = llama_get_kv_cache(ctx);
// clear previous kv_cache values (irrelevant for embeddings)
llama_kv_cache_clear(ctx);
llama_kv_cache_clear(kv);
// run model
LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);