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

llama : cont

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-01-13 14:56:52 +02:00
parent f78b396ee7
commit e4550fbafc
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
19 changed files with 128 additions and 79 deletions
Download patch file Download diff file Expand all files Collapse all files

14
examples/main/main.cpp
View file

@ -157,6 +157,8 @@ int main(int argc, char ** argv) {
return 1;
}
llama_kv_cache * kv = llama_get_kv_cache(ctx);
const llama_vocab * vocab = llama_model_get_vocab(model);
auto chat_templates = common_chat_templates_from_model(model, params.chat_template);
@ -328,7 +330,7 @@ int main(int argc, char ** argv) {
}
// remove any "future" tokens that we might have inherited from the previous session
llama_kv_cache_seq_rm(ctx, -1, n_matching_session_tokens, -1);
llama_kv_cache_seq_rm(kv, -1, n_matching_session_tokens, -1);
}
LOG_DBG("recalculate the cached logits (check): embd_inp.size() %zu, n_matching_session_tokens %zu, embd_inp.size() %zu, session_tokens.size() %zu\n",
@ -569,8 +571,8 @@ int main(int argc, char ** argv) {
LOG_DBG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
n_past, n_left, n_ctx, params.n_keep, n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep , params.n_keep + n_discard);
llama_kv_cache_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
llama_kv_cache_seq_rm (kv, 0, params.n_keep , params.n_keep + n_discard);
llama_kv_cache_seq_add(kv, 0, params.n_keep + n_discard, n_past, -n_discard);
n_past -= n_discard;
@ -593,9 +595,9 @@ int main(int argc, char ** argv) {
LOG_DBG("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n, (ga_i + ib*bd)/ga_n, (ga_i + ib*bd + ga_w)/ga_n);
LOG_DBG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i + ib*bd + ga_w, n_past + ib*bd, dd, ga_i + ib*bd + ga_w + dd, n_past + ib*bd + dd);
llama_kv_cache_seq_add(ctx, 0, ga_i, n_past, ib*bd);
llama_kv_cache_seq_div(ctx, 0, ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n);
llama_kv_cache_seq_add(ctx, 0, ga_i + ib*bd + ga_w, n_past + ib*bd, dd);
llama_kv_cache_seq_add(kv, 0, ga_i, n_past, ib*bd);
llama_kv_cache_seq_div(kv, 0, ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n);
llama_kv_cache_seq_add(kv, 0, ga_i + ib*bd + ga_w, n_past + ib*bd, dd);
n_past -= bd;