ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

sampling : when top-k <= 0 -> noop (#13173)

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-04-29 20:22:57 +03:00 • committed by GitHub
parent 5a63980117
commit d9d398f84f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

1
include/llama.h
View file

@ -1232,6 +1232,7 @@ extern "C" {
"will be removed in the future (see https://github.com/ggml-org/llama.cpp/pull/9896#discussion_r1800920915)");
/// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
/// Setting k <= 0 makes this a noop
LLAMA_API struct llama_sampler * llama_sampler_init_top_k (int32_t k);
/// @details Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751

3
src/llama-sampling.cpp
View file

@ -232,7 +232,7 @@ static void llama_sampler_top_k_impl(llama_token_data_array * cur_p, int32_t k)
// }
if (k <= 0) {
k = cur_p->size;
return;
}
k = std::min(k, (int) cur_p->size);
@ -298,6 +298,7 @@ static void llama_sampler_top_k_impl(llama_token_data_array * cur_p, int32_t k)
}
cur_p->sorted = true;
}
cur_p->size = k;
}