upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-10-02 19:07:25 -05:00
Code Issues Packages Projects Releases Wiki Activity

Resolve -1 to 1024 instead of ctx-len for samplers

Browse Source 
Because of backend-sampling we initialize samplers before the complete
llama_context is there. Therefore, we cannot infer the resolved context
length yet at the time we construct the samplers.
This commit is contained in:
Oliver Simons
2026-08-04 09:02:53 +03:00
committed by Georgi Gerganov
parent 935cad6497
commit 748eca633a
15 changed files with 78 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/sampling.cpp
+1 -6
View File
@@ -186,8 +186,7 @@ std::string common_params_sampling::print() const {
struct common_sampler * common_sampler_init(
const struct llama_model * model,
struct common_params_sampling & params,
int32_t n_ctx) {
struct common_params_sampling & params) {
if (!std::isfinite(params.penalty_repeat) ||
params.penalty_repeat <= 0.0f ||
!std::isfinite(1.0f/params.penalty_repeat)) {
@@ -199,10 +198,6 @@ struct common_sampler * common_sampler_init(
if (!std::isfinite(params.penalty_present)) {
throw std::invalid_argument("penalty_present must be finite");
}
if (params.penalty_last_n == -1) {
params.penalty_last_n = n_ctx > 0 ? n_ctx : llama_model_n_ctx_train(model);
}
const llama_vocab * vocab = llama_model_get_vocab(model);
llama_sampler_chain_params lparams = llama_sampler_chain_default_params();