CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

perplexity : fix ETA by warming up the model with an empty run

Browse Source
This commit is contained in:
Georgi Gerganov
2023-09-03 13:42:56 +03:00
parent 6519e9c99c
commit 8f429fa511
2 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
common/common.cpp
View File

@@ -752,6 +752,14 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
params.logit_bias[llama_token_eos(lctx)] = -INFINITY; params.logit_bias[llama_token_eos(lctx)] = -INFINITY;
} }
{
LOG("warming up the model with an empty run\n");
const std::vector<llama_token> tmp = { llama_token_bos(lctx), };
llama_eval(lctx, tmp.data(), tmp.size(), 0, params.n_threads);
llama_reset_timings(lctx);
}
return std::make_tuple(model, lctx); return std::make_tuple(model, lctx);
} }

8
examples/main/main.cpp
View File

@@ -492,14 +492,6 @@ int main(int argc, char ** argv) {
std::vector<llama_token> embd; std::vector<llama_token> embd;
std::vector<llama_token> embd_guidance; std::vector<llama_token> embd_guidance;
{
LOG("warming up the model with an empty run\n");
const std::vector<llama_token> tmp = { llama_token_bos(ctx), };
llama_eval(ctx, tmp.data(), tmp.size(), 0, params.n_threads);
llama_reset_timings(ctx);
}
while ((n_remain != 0 && !is_antiprompt) || params.interactive) { while ((n_remain != 0 && !is_antiprompt) || params.interactive) {
// predict // predict
if (embd.size() > 0) { if (embd.size() > 0) {