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

set cparams.n_parallel to the number of sequences

Browse source
This commit is contained in:
slaren 2024-03-09 03:42:37 +01:00
parent 7d999555b7
commit ac07f7d0f7
2 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
examples/perplexity/perplexity.cpp
View file

@ -1822,7 +1822,9 @@ int main(int argc, char ** argv) {
const bool ppl = !params.hellaswag && !params.winogrande && !params.multiple_choice && !params.kl_divergence;
if (ppl) {
int32_t n_kv = std::max(1, params.n_batch / n_ctx) * n_ctx;
int n_seq = std::max(1, params.n_batch / n_ctx);
int32_t n_kv = n_seq * n_ctx;
params.n_parallel = n_seq;
params.n_ctx = n_kv;
params.n_batch = std::min(params.n_batch, n_kv);
} else {

2
llama.cpp
View file

@ -8943,7 +8943,7 @@ static int llama_decode_internal(
}
}
#ifndef NDEBUG
logits_valid[i] = batch.logits[i] == 1;
logits_valid[i] = batch.logits[i] != 0;
#endif
}
} else if (lctx.logits_all) {