tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 12:05:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

examples : do not assume BOS when shifting context (#5622)

Browse Source
This commit is contained in:
Jared Van Bortel
2024-02-21 10:33:54 -05:00
committed by GitHub
parent 5022cf242d
commit 89febfed93
2 changed files with 14 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/main/main.cpp
View File

@ -334,6 +334,8 @@ int main(int argc, char ** argv) {
// number of tokens to keep when resetting context
if (params.n_keep < 0 || params.n_keep > (int) embd_inp.size() || params.instruct || params.chatml) {
params.n_keep = (int)embd_inp.size();
} else {
params.n_keep += add_bos; // always keep the BOS token
}
// prefix & suffix for instruct mode
@ -383,8 +385,8 @@ int main(int argc, char ** argv) {
}
}
if (params.n_keep > 0) {
LOG_TEE("%s: static prompt based on n_keep: '", __func__);
if (params.n_keep > add_bos) {
LOG_TEE("%s: static prompt based on n_keep: '", __func__);
for (int i = 0; i < params.n_keep; i++) {
LOG_TEE("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str());
}
@ -540,14 +542,14 @@ int main(int argc, char ** argv) {
break;
}
const int n_left = n_past - params.n_keep - 1;
const int n_left = n_past - params.n_keep;
const int n_discard = n_left/2;
LOG("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 + 1 , params.n_keep + n_discard + 1);
llama_kv_cache_seq_shift(ctx, 0, params.n_keep + 1 + n_discard, n_past, -n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep , params.n_keep + n_discard);
llama_kv_cache_seq_shift(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
n_past -= n_discard;