vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-09 18:44:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add args -k and --force

Browse source
This commit is contained in:
Zonghang Li 2025-03-11 20:44:36 +04:00
parent 9cbdf01645
commit bcfdace59b
5 changed files with 27 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

14
common/arg.cpp
View file

@ -737,6 +737,20 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
params.gpu_mem = value; // in GiB
}
).set_env("LLAMA_ARG_CUDA_MEM"));
add_opt(llama_arg(
{"-k", "--n-cycles"}, "N",
format("number of cycles to output one token (default: %d)", params.n_cycles),
[](gpt_params & params, int value) {
params.n_cycles = value;
}
).set_env("LLAMA_ARG_K"));
add_opt(llama_arg(
{"--force"},
format("force to start prefetching after computation (default: %s)", params.force ? "true" : "false"),
[](gpt_params & params) {
params.force = true;
}
).set_env("LLAMA_ARG_FORCE"));
// #ifdef GGML_USE_METAL
// // warn: if the output layer weights are not kept in metal shared memory, its mmap-ed weight data
// // could be released by the OS and reloaded repeatedly, which causes additional disk I/O latency.