Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.devops/vulkan.Dockerfile
#	.github/workflows/build.yml
#	.github/workflows/server.yml
#	common/common.cpp
#	examples/batched/README.md
#	ggml/CMakeLists.txt
#	ggml/src/CMakeLists.txt
#	ggml/src/ggml-cann/ggml-cann.cpp
#	ggml/src/ggml-cpu/CMakeLists.txt
#	ggml/src/ggml-cpu/arch-fallback.h
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	scripts/sync-ggml.last
#	src/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tools/server/CMakeLists.txt
This commit is contained in:
Concedo 2025-11-25 16:38:07 +08:00
commit 724763fdec
55 changed files with 5248 additions and 3753 deletions

View file

@ -1234,6 +1234,7 @@ 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(
@ -1263,6 +1264,7 @@ 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(
@ -1270,6 +1272,7 @@ 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(
@ -1277,6 +1280,7 @@ 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(
@ -1284,6 +1288,7 @@ 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(
@ -1298,6 +1303,7 @@ 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(
@ -1305,6 +1311,7 @@ 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(
@ -1323,6 +1330,7 @@ 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(
@ -1330,6 +1338,7 @@ 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(
@ -1427,6 +1436,7 @@ 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(
@ -1434,6 +1444,7 @@ 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(
@ -1441,6 +1452,7 @@ 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(