root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 11:24:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

common : add missing env var for speculative (#10801)

Browse Source
This commit is contained in:
Xuan Son Nguyen 2024-12-12 16:57:32 +01:00 committed by GitHub
parent 5555c0c1f6
commit 9fdb124304
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
common/arg.cpp
View File

@ -2083,35 +2083,35 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
[](common_params & params, int value) { [](common_params & params, int value) {
params.speculative.n_max = value; params.speculative.n_max = value;
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_DRAFT_MAX"));
add_opt(common_arg( add_opt(common_arg(
{"--draft-min", "--draft-n-min"}, "N", {"--draft-min", "--draft-n-min"}, "N",
string_format("minimum number of draft tokens to use for speculative decoding (default: %d)", params.speculative.n_min), string_format("minimum number of draft tokens to use for speculative decoding (default: %d)", params.speculative.n_min),
[](common_params & params, int value) { [](common_params & params, int value) {
params.speculative.n_min = value; params.speculative.n_min = value;
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_DRAFT_MIN"));
add_opt(common_arg( add_opt(common_arg(
{"--draft-p-split"}, "P", {"--draft-p-split"}, "P",
string_format("speculative decoding split probability (default: %.1f)", (double)params.speculative.p_split), string_format("speculative decoding split probability (default: %.1f)", (double)params.speculative.p_split),
[](common_params & params, const std::string & value) { [](common_params & params, const std::string & value) {
params.speculative.p_split = std::stof(value); params.speculative.p_split = std::stof(value);
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}).set_env("LLAMA_ARG_DRAFT_P_SPLIT"));
add_opt(common_arg( add_opt(common_arg(
{"--draft-p-min"}, "P", {"--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: %.1f)", (double)params.speculative.p_min),
[](common_params & params, const std::string & value) { [](common_params & params, const std::string & value) {
params.speculative.p_min = std::stof(value); params.speculative.p_min = std::stof(value);
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_DRAFT_P_MIN"));
add_opt(common_arg( add_opt(common_arg(
{"-cd", "--ctx-size-draft"}, "N", {"-cd", "--ctx-size-draft"}, "N",
string_format("size of the prompt context for the draft model (default: %d, 0 = loaded from model)", params.speculative.n_ctx), string_format("size of the prompt context for the draft model (default: %d, 0 = loaded from model)", params.speculative.n_ctx),
[](common_params & params, int value) { [](common_params & params, int value) {
params.speculative.n_ctx = value; params.speculative.n_ctx = value;
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CTX_SIZE_DRAFT"));
add_opt(common_arg( add_opt(common_arg(
{"-devd", "--device-draft"}, "<dev1,dev2,..>", {"-devd", "--device-draft"}, "<dev1,dev2,..>",
"comma-separated list of devices to use for offloading the draft model (none = don't offload)\n" "comma-separated list of devices to use for offloading the draft model (none = don't offload)\n"
@ -2131,14 +2131,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
fprintf(stderr, "warning: consult docs/build.md for compilation instructions\n"); fprintf(stderr, "warning: consult docs/build.md for compilation instructions\n");
} }
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_N_GPU_LAYERS_DRAFT"));
add_opt(common_arg( add_opt(common_arg(
{"-md", "--model-draft"}, "FNAME", {"-md", "--model-draft"}, "FNAME",
"draft model for speculative decoding (default: unused)", "draft model for speculative decoding (default: unused)",
[](common_params & params, const std::string & value) { [](common_params & params, const std::string & value) {
params.speculative.model = value; params.speculative.model = value;
} }
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_MODEL_DRAFT"));
return ctx_arg; return ctx_arg;
} }