tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-15 04:33:06 -04:00
Code Issues Packages Projects Releases Wiki Activity

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
7 changed files with 202 additions and 218 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/speculative/speculative.cpp
View File

@@ -185,8 +185,6 @@ int main(int argc, char ** argv) {
// target model sampling context (reuse the llama_context's sampling instance)
struct common_sampler * smpl = common_sampler_init(model_tgt, params.sparams);
struct llama_sampler * softmax = llama_sampler_init_softmax();
// draft sequence data
std::vector<seq_draft> drafts(n_seq_dft);
@@ -629,7 +627,6 @@ int main(int argc, char ** argv) {
common_sampler_free(drafts[s].smpl);
}
llama_sampler_free(softmax);
llama_batch_free(batch_dft);
llama_free(ctx_tgt);