Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-18 16:55:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/workflows/check-vendor.yml
#	.github/workflows/close-issue.yml
#	.github/workflows/editorconfig.yml
#	.github/workflows/gguf-publish.yml
#	.github/workflows/labeler.yml
#	.github/workflows/pre-tokenizer-hashes.yml
#	.github/workflows/python-check-requirements.yml
#	.github/workflows/python-lint.yml
#	.github/workflows/python-type-check.yml
#	.github/workflows/server.yml
#	.github/workflows/update-ops-docs.yml
#	README.md
#	docs/build.md
#	examples/model-conversion/scripts/utils/perplexity-gen.sh
#	examples/model-conversion/scripts/utils/perplexity-run-simple.sh
#	examples/model-conversion/scripts/utils/perplexity-run.sh
#	examples/model-conversion/scripts/utils/quantize.sh
#	examples/model-conversion/scripts/utils/run-embedding-server.sh
#	ggml/src/ggml-cpu/ggml-cpu.c
#	ggml/src/ggml-hexagon/htp/flash-attn-ops.c
#	ggml/src/ggml-opencl/CMakeLists.txt
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-opencl/kernels/cvt.cl
#	ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32.cl
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	scripts/compare-llama-bench.py
#	tests/test-backend-ops.cpp
#	tests/test-gguf.cpp
#	tools/cli/README.md
#	tools/completion/README.md
#	tools/server/README.md
This commit is contained in:
Concedo
2026-01-27 23:06:13 +08:00
parent 3a84623ea5 68ac3acb43
commit f6ece6fd37
40 changed files with 3108 additions and 477 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/arg.cpp
+25 -21
View File
@@ -1234,6 +1234,10 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
string_format("size of the prompt context (default: %d, 0 = loaded from model)", params.n_ctx),
[](common_params & params, int value) {
params.n_ctx = value;
if (value == 0) {
// disable context reduction in llama_params_fit if the user explicitly requests the full context size:
params.fit_params_min_ctx = UINT32_MAX;
}
}
).set_env("LLAMA_ARG_CTX_SIZE"));
add_opt(common_arg(
@@ -1576,7 +1580,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--temp"}, "N",
string_format("temperature (default: %.1f)", (double)params.sampling.temp),
string_format("temperature (default: %.2f)", (double)params.sampling.temp),
[](common_params & params, const std::string & value) {
params.sampling.temp = std::stof(value);
params.sampling.temp = std::max(params.sampling.temp, 0.0f);
@@ -1593,7 +1597,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam().set_env("LLAMA_ARG_TOP_K"));
add_opt(common_arg(
{"--top-p"}, "N",
string_format("top-p sampling (default: %.1f, 1.0 = disabled)", (double)params.sampling.top_p),
string_format("top-p sampling (default: %.2f, 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;
@@ -1601,7 +1605,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--min-p"}, "N",
string_format("min-p sampling (default: %.1f, 0.0 = disabled)", (double)params.sampling.min_p),
string_format("min-p sampling (default: %.2f, 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;
@@ -1609,14 +1613,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--top-nsigma"}, "N",
string_format("top-n-sigma sampling (default: %.1f, -1.0 = disabled)", params.sampling.top_n_sigma),
string_format("top-n-sigma sampling (default: %.2f, -1.0 = disabled)", params.sampling.top_n_sigma),
[](common_params & params, const std::string & value) {
params.sampling.top_n_sigma = std::stof(value);
}
).set_sparam());
add_opt(common_arg(
{"--xtc-probability"}, "N",
string_format("xtc probability (default: %.1f, 0.0 = disabled)", (double)params.sampling.xtc_probability),
string_format("xtc probability (default: %.2f, 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;
@@ -1624,7 +1628,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--xtc-threshold"}, "N",
string_format("xtc threshold (default: %.1f, 1.0 = disabled)", (double)params.sampling.xtc_threshold),
string_format("xtc threshold (default: %.2f, 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;
@@ -1632,7 +1636,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--typical"}, "N",
string_format("locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)params.sampling.typ_p),
string_format("locally typical sampling, parameter p (default: %.2f, 1.0 = disabled)", (double)params.sampling.typ_p),
[](common_params & params, const std::string & value) {
params.sampling.typ_p = std::stof(value);
}
@@ -1651,7 +1655,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--repeat-penalty"}, "N",
string_format("penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)params.sampling.penalty_repeat),
string_format("penalize repeat sequence of tokens (default: %.2f, 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;
@@ -1659,21 +1663,21 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--presence-penalty"}, "N",
string_format("repeat alpha presence penalty (default: %.1f, 0.0 = disabled)", (double)params.sampling.penalty_present),
string_format("repeat alpha presence penalty (default: %.2f, 0.0 = disabled)", (double)params.sampling.penalty_present),
[](common_params & params, const std::string & value) {
params.sampling.penalty_present = std::stof(value);
}
).set_sparam());
add_opt(common_arg(
{"--frequency-penalty"}, "N",
string_format("repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)", (double)params.sampling.penalty_freq),
string_format("repeat alpha frequency penalty (default: %.2f, 0.0 = disabled)", (double)params.sampling.penalty_freq),
[](common_params & params, const std::string & value) {
params.sampling.penalty_freq = std::stof(value);
}
).set_sparam());
add_opt(common_arg(
{"--dry-multiplier"}, "N",
string_format("set DRY sampling multiplier (default: %.1f, 0.0 = disabled)", (double)params.sampling.dry_multiplier),
string_format("set DRY sampling multiplier (default: %.2f, 0.0 = disabled)", (double)params.sampling.dry_multiplier),
[](common_params & params, const std::string & value) {
params.sampling.dry_multiplier = std::stof(value);
}
@@ -1754,14 +1758,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--dynatemp-range"}, "N",
string_format("dynamic temperature range (default: %.1f, 0.0 = disabled)", (double)params.sampling.dynatemp_range),
string_format("dynamic temperature range (default: %.2f, 0.0 = disabled)", (double)params.sampling.dynatemp_range),
[](common_params & params, const std::string & value) {
params.sampling.dynatemp_range = std::stof(value);
}
).set_sparam());
add_opt(common_arg(
{"--dynatemp-exp"}, "N",
string_format("dynamic temperature exponent (default: %.1f)", (double)params.sampling.dynatemp_exponent),
string_format("dynamic temperature exponent (default: %.2f)", (double)params.sampling.dynatemp_exponent),
[](common_params & params, const std::string & value) {
params.sampling.dynatemp_exponent = std::stof(value);
}
@@ -1777,7 +1781,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--mirostat-lr"}, "N",
string_format("Mirostat learning rate, parameter eta (default: %.1f)", (double)params.sampling.mirostat_eta),
string_format("Mirostat learning rate, parameter eta (default: %.2f)", (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;
@@ -1785,7 +1789,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_sparam());
add_opt(common_arg(
{"--mirostat-ent"}, "N",
string_format("Mirostat target entropy, parameter tau (default: %.1f)", (double)params.sampling.mirostat_tau),
string_format("Mirostat target entropy, parameter tau (default: %.2f)", (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;
@@ -1919,28 +1923,28 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_env("LLAMA_ARG_YARN_ORIG_CTX"));
add_opt(common_arg(
{"--yarn-ext-factor"}, "N",
string_format("YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor),
string_format("YaRN: extrapolation mix factor (default: %.2f, 0.0 = full interpolation)", (double)params.yarn_ext_factor),
[](common_params & params, const std::string & value) {
params.yarn_ext_factor = std::stof(value);
}
).set_env("LLAMA_ARG_YARN_EXT_FACTOR"));
add_opt(common_arg(
{"--yarn-attn-factor"}, "N",
string_format("YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor),
string_format("YaRN: scale sqrt(t) or attention magnitude (default: %.2f)", (double)params.yarn_attn_factor),
[](common_params & params, const std::string & value) {
params.yarn_attn_factor = std::stof(value);
}
).set_env("LLAMA_ARG_YARN_ATTN_FACTOR"));
add_opt(common_arg(
{"--yarn-beta-slow"}, "N",
string_format("YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow),
string_format("YaRN: high correction dim or alpha (default: %.2f)", (double)params.yarn_beta_slow),
[](common_params & params, const std::string & value) {
params.yarn_beta_slow = std::stof(value);
}
).set_env("LLAMA_ARG_YARN_BETA_SLOW"));
add_opt(common_arg(
{"--yarn-beta-fast"}, "N",
string_format("YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast),
string_format("YaRN: low correction dim or beta (default: %.2f)", (double)params.yarn_beta_fast),
[](common_params & params, const std::string & value) {
params.yarn_beta_fast = std::stof(value);
}
@@ -3334,14 +3338,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_DRAFT_MIN"));
add_opt(common_arg(
{"--draft-p-split"}, "P",
string_format("speculative decoding split probability (default: %.1f)", (double)params.speculative.p_split),
string_format("speculative decoding split probability (default: %.2f)", (double)params.speculative.p_split),
[](common_params & params, const std::string & value) {
params.speculative.p_split = std::stof(value);
}
).set_examples({LLAMA_EXAMPLE_SPECULATIVE}).set_env("LLAMA_ARG_DRAFT_P_SPLIT"));
add_opt(common_arg(
{"--draft-p-min"}, "P",
string_format("minimum speculative decoding probability (greedy) (default: %.1f)", (double)params.speculative.p_min),
string_format("minimum speculative decoding probability (greedy) (default: %.2f)", (double)params.speculative.p_min),
[](common_params & params, const std::string & value) {
params.speculative.p_min = std::stof(value);
}