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

sampling : refactor init to use llama_sampling_params

Browse source
This commit is contained in:
Georgi Gerganov 2023-10-20 14:58:20 +03:00
parent 8cf19d60dc
commit cd1e937821
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
12 changed files with 110 additions and 142 deletions
Download patch file Download diff file Expand all files Collapse all files

8
examples/speculative/speculative.cpp
View file

@ -112,16 +112,16 @@ int main(int argc, char ** argv) {
bool has_eos = false;
// target model sampling context
struct llama_sampling_context * ctx_sampling = llama_sampling_init(params);
struct llama_sampling_context * ctx_sampling = llama_sampling_init(params.sparams);
// draft sequence data
std::vector<seq_draft> drafts(n_seq_dft);
params.grammar.clear(); // the draft samplers will copy the target sampler's grammar
params.sampling_params.temp = std::max(0.01f, params.sampling_params.temp);
params.sparams.grammar.clear(); // the draft samplers will copy the target sampler's grammar
params.sparams.temp = std::max(0.01f, params.sparams.temp);
for (int s = 0; s < n_seq_dft; ++s) {
drafts[s].ctx_sampling = llama_sampling_init(params);
drafts[s].ctx_sampling = llama_sampling_init(params.sparams);
}
llama_batch batch_dft = llama_batch_init(params.n_ctx, 0, 1);