sleepy/llama.cpp
1
0
Fork 0
Code Issues 12 Pull Requests Actions 113 Packages Projects Releases Wiki Activity

llama: introduce support for model-embedded sampling parameters (#17120)

Browse Source
This commit is contained in:
Aaron Teo
2025-11-25 09:56:07 +08:00
committed by GitHub
parent 3d07caa99b
commit 877566d512
10 changed files with 293 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/arg.cpp
+12
View File
@@ -1232,6 +1232,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
[](common_params & params, const std::string & value) {
const auto sampler_names = string_split<std::string>(value, ';');
params.sampling.samplers = common_sampler_types_from_names(sampler_names, true);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_SAMPLERS;
}
).set_sparam());
add_opt(common_arg(
@@ -1261,6 +1262,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
[](common_params & params, const std::string & value) {
params.sampling.temp = std::stof(value);
params.sampling.temp = std::max(params.sampling.temp, 0.0f);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TEMP;
}
).set_sparam());
add_opt(common_arg(
@@ -1268,6 +1270,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("top-k sampling (default: %d, 0 = disabled)", params.sampling.top_k),
[](common_params & params, int value) {
params.sampling.top_k = value;
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TOP_K;
}
).set_sparam());
add_opt(common_arg(
@@ -1275,6 +1278,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("top-p sampling (default: %.1f, 1.0 = disabled)", (double)params.sampling.top_p),
[](common_params & params, const std::string & value) {
params.sampling.top_p = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TOP_P;
}
).set_sparam());
add_opt(common_arg(
@@ -1282,6 +1286,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("min-p sampling (default: %.1f, 0.0 = disabled)", (double)params.sampling.min_p),
[](common_params & params, const std::string & value) {
params.sampling.min_p = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIN_P;
}
).set_sparam());
add_opt(common_arg(
@@ -1296,6 +1301,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("xtc probability (default: %.1f, 0.0 = disabled)", (double)params.sampling.xtc_probability),
[](common_params & params, const std::string & value) {
params.sampling.xtc_probability = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_XTC_PROBABILITY;
}
).set_sparam());
add_opt(common_arg(
@@ -1303,6 +1309,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("xtc threshold (default: %.1f, 1.0 = disabled)", (double)params.sampling.xtc_threshold),
[](common_params & params, const std::string & value) {
params.sampling.xtc_threshold = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_XTC_THRESHOLD;
}
).set_sparam());
add_opt(common_arg(
@@ -1321,6 +1328,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
}
params.sampling.penalty_last_n = value;
params.sampling.n_prev = std::max(params.sampling.n_prev, params.sampling.penalty_last_n);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_LAST_N;
}
).set_sparam());
add_opt(common_arg(
@@ -1328,6 +1336,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)params.sampling.penalty_repeat),
[](common_params & params, const std::string & value) {
params.sampling.penalty_repeat = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_REPEAT;
}
).set_sparam());
add_opt(common_arg(
@@ -1425,6 +1434,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
"(default: %d, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)", params.sampling.mirostat),
[](common_params & params, int value) {
params.sampling.mirostat = value;
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT;
}
).set_sparam());
add_opt(common_arg(
@@ -1432,6 +1442,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("Mirostat learning rate, parameter eta (default: %.1f)", (double)params.sampling.mirostat_eta),
[](common_params & params, const std::string & value) {
params.sampling.mirostat_eta = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_ETA;
}
).set_sparam());
add_opt(common_arg(
@@ -1439,6 +1450,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("Mirostat target entropy, parameter tau (default: %.1f)", (double)params.sampling.mirostat_tau),
[](common_params & params, const std::string & value) {
params.sampling.mirostat_tau = std::stof(value);
params.sampling.user_sampling_config |= common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_TAU;
}
).set_sparam());
add_opt(common_arg(