tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-29 04:35:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

speculative : change default p_accept to 0.5 + CLI args (#3919)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2023-11-03 09:41:17 +02:00
parent 05816027d6
commit 8f961abdc4
3 changed files with 25 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
examples/speculative/speculative.cpp
View File

@ -37,9 +37,11 @@ int main(int argc, char ** argv) {
// max number of parallel drafting sequences (i.e. tree branches)
const int n_seq_dft = params.n_parallel;
// TODO: make this configurable
const float p_accept = 0.80f;
const float p_split = 0.10f;
// probability threshold for accepting a token from the draft model
const float p_accept = params.p_accept;
// probability threshold for splitting a draft branch (only for n_seq_dft > 1)
const float p_split = params.p_split;
#ifndef LOG_DISABLE_LOGS
log_set_target(log_filename_generator("speculative", "log"));