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

llama : default sampling changes + greedy update (#9897)

Browse source 
* llama : deprecate softmax sampler + fix dist sampler

ggml-ci

* tests : replace macros with functions

ggml-ci

* sampling : change temperature sampler logic

For t <= 0.0f, keep the max logit intact and set the rest to -inf

* cont : no need for special "greedy" logic

top-k == 1 is the same

* tests : init prob correctly

* llama : handle temp <= 0.0 in the temp_ext sampler too

ggml-ci

* cont : avoid extra loop in temperature sampler for sub-zero temp

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-10-21 09:46:40 +03:00 • committed by GitHub
parent bc21975084
commit 55e47786e3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 202 additions and 218 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/save-load-state/save-load-state.cpp
View file

@ -42,7 +42,6 @@ int main(int argc, char ** argv) {
llama_sampler * smpl = llama_sampler_chain_init(sparams);
llama_sampler_chain_add(smpl, llama_sampler_init_softmax());
llama_sampler_chain_add(smpl, llama_sampler_init_dist(params.sparams.seed));
// tokenize prompt
@ -107,7 +106,6 @@ int main(int argc, char ** argv) {
llama_sampler * smpl2 = llama_sampler_chain_init(sparams);
llama_sampler_chain_add(smpl2, llama_sampler_init_softmax());
llama_sampler_chain_add(smpl2, llama_sampler_init_dist(params.sparams.seed));
printf("\nsecond run: %s", params.prompt.c_str());
@ -171,7 +169,6 @@ int main(int argc, char ** argv) {
llama_sampler * smpl3 = llama_sampler_chain_init(sparams);
llama_sampler_chain_add(smpl3, llama_sampler_init_softmax());
llama_sampler_chain_add(smpl3, llama_sampler_init_dist(params.sparams.seed));
printf("\nsingle seq run: %s", params.prompt.c_str());