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

moved reset to end of llama_decode_internal

Browse source
This commit is contained in:
Alan Gray 2024-04-26 10:24:02 -07:00
parent a2beaffec8
commit 34847caa9a
1 changed files with 4 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

9
llama.cpp
View file

@ -11205,6 +11205,10 @@ static int llama_decode_internal(
}
}
// Reset state for the next token before backend sync, to allow the CPU activities in the reset to
// overlap with device computation.
ggml_backend_sched_reset(lctx.sched);
return 0;
}
@ -16773,11 +16777,6 @@ float * llama_get_logits(struct llama_context * ctx) {
float * llama_get_logits_ith(struct llama_context * ctx, int32_t i) {
int32_t j = -1;
// Reset state for the next run before the following backend sync,
// to allow the CPU activities in the reset to overlap with device computation.
ggml_backend_sched_reset(ctx->sched);
llama_synchronize(ctx);
try {