CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

sampling : make top_n_sigma no-op at <=0 or a single candidate (#13345)

Browse Source
This commit is contained in:
DocShotgun
2025-05-06 13:36:24 -07:00
committed by GitHub
parent 91a86a6f35
commit ffc727203a
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-sampling.cpp
View File

@@ -1750,7 +1750,7 @@ static const char * llama_sampler_top_n_sigma_name(const struct llama_sampler *
static void llama_sampler_top_n_sigma_apply(struct llama_sampler * smpl, llama_token_data_array * cur_p) {
const auto * ctx = (llama_sampler_top_n_sigma *) smpl->ctx;
if (ctx->n < 0.0f) {
if (ctx->n <= 0.0f || cur_p->size <= 1) {
return;
}

2
tests/test-sampling.cpp
View File

@@ -360,7 +360,7 @@ int main(void) {
test_dry({0.2f, 0.2f, 0.2f, 0.2f, 0.2f}, {0, 1, 2, 3, 4, 0, 1}, {0.2f, 0.2f, 0.2f, 0.2f, 0.2f}, 1.0f, 1.1f, 4, 7, {});
test_top_n_sigma({0.1f, 0.2f, 0.3f, 0.4f}, {0.571429f, 0.428571f, 0.0f, 0.0f}, 1.00f);
test_top_n_sigma({0.1f, 0.2f, 0.3f, 0.4f}, {1.0f, 0.0f, 0.0f, 0.0f}, 0.00f);
test_top_n_sigma({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f, 0.3f, 0.2f, 0.1f}, 0.00f); // top_n_sigma == 0 now represents a no-op rather than greedy decoding as of PR#13345
test_top_n_sigma({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f, 0.3f, 0.2f, 0.1f}, 3.00f);
test_sampler_queue(10000, "k", 10000, 1.0f, 1.0f);