Is there a way to silence log/debug output?

[oelna]

When I run the program, there is a lot of output other than the model response, such as

main: build = 0 (unknown)
main: seed  = 1686229235
llama.cpp: loading model from /models/selfee-13b.ggmlv3.q4_0.bin
llama_model_load_internal: format     = ggjt v3 (latest)
llama_model_load_internal: n_vocab    = 32001
llama_model_load_internal: n_ctx      = 512
llama_model_load_internal: n_embd     = 5120
…
llama_print_timings: prompt eval time =  9715.31 ms /     5 tokens ( 1943.06 ms per token)
llama_print_timings:        eval time = 28065.27 ms /   337 runs   (   83.28 ms per token)
llama_print_timings:       total time = 38073.44 ms

Is there a way to prevent this output of log messages and stats, such as a verbosity level I can set as a parameter? Or a --quiet or --silent mode?

I would like to use the output in other contexts, such as shell scripts, but parsing out the log messages seems like the wrong way to go about it. I'm grateful for any tips or hints.

[j-f1]

If you redirect stderr you should only see the output (e.g. ./main 2>/dev/null)

[oelna]

Amazingly simple solution. Thank you!

[waffletower]

llama-cli still litters the current directory with a main.log file while using this STDERR redirect solution. I prefer to use the --log-disable flag:

llama-cli --log-disable --prompt "I have nothing to say and I am saying it."

[hasaranga]

what can I use on windows?

[rhatdan]

We are trying to build github.com/containers/ramalama which is wriapping llama-cli, and want to block the debugging noice, and are currently doing the 2> /dev/null thing, but this also blocks real errors from showing up.

When I tried the --log-disable, I see nothing almost like stdout and stderr are blocked, command just freezes.

llama-cli -m /home/dwalsh/.local/share/ramalama/models/ollama/tinyllama:latest --in-prefix '' --in-suffix '' -p 'You are a helpful assistant' -c 2048 -cnv --log-disable

But:

$ llama-cli -m /home/dwalsh/.local/share/ramalama/models/ollama/tinyllama:latest --in-prefix '' --in-suffix '' -p 'You are a helpful assistant' -c 2048 -cnv 2>/dev/null 
.....................................................................................
main: interactive mode on.

== Running in interactive mode. ==
 - Press Ctrl+C to interject at any time.
 - Press Return to return control to the AI.
 - To return control without starting a new line, end your input with '/'.
 - If you want to submit another line, end your input with '\'.

 You are a helpful assistant
> 

Issue we have is:
llama-cli -m /home/dwalsh/.local/share/ramalama/models/ollama/missing --in-prefix '' --in-suffix '' -p 'You are a helpful assistant' -c 2048 -cnv 2>/dev/null

Does not show the error:

llama_model_load: error loading model: llama_model_loader: failed to load model from /home/dwalsh/.local/share/ramalama/models/ollama/missing

[ericcurtin]

This was briefly discussed here when the behaviour changed #9418

[missionsave]

c++:

static void llama_log_callback_null(ggml_log_level level, const char * text, void * user_data) { (void) level; (void) text; (void) user_data; }

then call
llama_log_set(llama_log_callback_null, NULL);

[RhinoDevel]

Maybe an alternative: common_log_pause(common_log_main());

[missionsave]

llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
    (void) level;
    (void) text;
    (void) user_data;
}, NULL);