upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-07 02:20:48 -05:00
Code Issues Packages Projects Releases Wiki Activity

sampling : add min-p backend sampler

Browse Source
This commit is contained in:
Daniel Bevenius
2025-11-26 10:50:58 +01:00
parent f23b306cc5
commit b45d504e70
4 changed files with 188 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
common/sampling.cpp
View File

@@ -173,6 +173,7 @@ static bool sampler_backend_supported(enum common_sampler_type type) {
switch (type) {
case COMMON_SAMPLER_TYPE_TOP_K:
case COMMON_SAMPLER_TYPE_TEMPERATURE:
case COMMON_SAMPLER_TYPE_MIN_P:
return true;
default:
return false;
@@ -325,6 +326,12 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
}
backend_idx++;
break;
case COMMON_SAMPLER_TYPE_MIN_P:
if (params.min_p > 0.0f) {
llama_sampler_chain_add(result->backend_chain, llama_sampler_backend_init_min_p(params.min_p));
}
backend_idx++;
break;
default:
GGML_ASSERT(false && "unsupported backend sampler");
}
@@ -468,6 +475,12 @@ struct llama_sampler * common_sampler_backend_init(const struct llama_model * mo
}
backend_idx++;
break;
case COMMON_SAMPLER_TYPE_MIN_P:
if (params.min_p > 0.0f) {
llama_sampler_chain_add(chain, llama_sampler_backend_init_min_p(params.min_p));
}
backend_idx++;
break;
default:
GGML_ASSERT(false && "unsupported backend sampler");
}