tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-02 13:25:43 +00:00
Code Issues Packages Projects Releases Wiki Activity

sampling : min-p should always return at least one token

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-05-27 11:24:40 +03:00
parent 4f81b33e32
commit fe12a5d47e
2 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-sampling.cpp
View File

@ -798,7 +798,7 @@ static void llama_sampler_min_p_apply(struct llama_sampler * smpl, llama_token_d
}
// if we have enough values the operation was a success
if (filtered_tokens.size() >= ctx->min_keep) {
if (!filtered_tokens.empty() && filtered_tokens.size() >= ctx->min_keep) {
memcpy(cur_p->data, filtered_tokens.data(), filtered_tokens.size()*sizeof(llama_token_data));
cur_p->size = filtered_tokens.size();
min_p_applied = true;

3
tests/test-sampling.cpp
View File

@ -109,7 +109,7 @@ static void test_min_p(const std::vector<float> & probs, const std::vector<float
sampler_tester tester(probs, probs_expected);
DUMP(&tester.cur_p);
tester.apply(llama_sampler_init_min_p(p, 1));
tester.apply(llama_sampler_init_min_p(p, 0));
tester.apply(llama_sampler_init_dist (0));
DUMP(&tester.cur_p);
@ -332,6 +332,7 @@ int main(void) {
test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.7f, 0.3f/0.7f}, 0.74f);
test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 0.76f);
test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 1.00f);
test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 1.05f);
printf("XTC should:\n");
test_xtc({0.4f, 0.3f, 0.2f, 0.1f}, {0.1f}, 0.99f, 0.09f);