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

fixed off by one error when context shifting

Browse source
This commit is contained in:
l3utterfly 2024-04-26 17:53:08 +09:00
parent 784e11dea1
commit ff2a6e845b
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
examples/main/main.cpp
View file

@ -544,7 +544,7 @@ int main(int argc, char ** argv) {
// if we run out of context:
// - take the n_keep first tokens from the original prompt (via n_past)
// - take half of the last (n_ctx - n_keep) tokens and recompute the logits in batches
if (n_past + (int) embd.size() + std::max<int>(0, guidance_offset) > n_ctx) {
if (n_past + (int) embd.size() + std::max<int>(0, guidance_offset) >= n_ctx) {
if (params.n_predict == -2) {
LOG_TEE("\n\n%s: context full and n_predict == -%d => stopping\n", __func__, params.n_predict);
break;