mirror of
https://github.com/Lizonghang/prima.cpp.git
synced 2025-09-10 12:44:34 +00:00
add mmap prefetch and unloading
This commit is contained in:
parent
ba5117581e
commit
c97ea10617
7 changed files with 161 additions and 11 deletions
|
@ -696,6 +696,13 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
|||
params.next_node_ip = value;
|
||||
}
|
||||
).set_env("LLAMA_ARG_NEXT_NODE_IP"));
|
||||
add_opt(llama_arg(
|
||||
{"--unload", "--unload-weight"},
|
||||
format("whether to unload layer weights after use (default: %s)", params.unload ? "true" : "false"),
|
||||
[](gpt_params & params) {
|
||||
params.unload = true;
|
||||
}
|
||||
).set_env("LLAMA_ARG_UNLOAD"));
|
||||
add_opt(llama_arg(
|
||||
{"-n", "--predict", "--n-predict"}, "N",
|
||||
format("number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict),
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue