Can't seem to turn off logs

[vlovich]

I have the following params set which the docs indicate should turn off stdout/stderr noise:

        params.set_debug_mode(false);
        params.set_print_special(false);
        params.set_print_progress(false);
        params.set_print_realtime(false);
        params.set_print_timestamps(false);

but for some reason I'm still seeing output like the following on my terminal:

whisper_full_with_state: decoder 0 completed
whisper_full_with_state: decoder  0: score = -0.88475, result_len =  10, avg_logprobs = -0.88475, entropy =  2.16396
whisper_full_with_state: best decoder = 0
seek = 500, seek_delta = 500

whisper_full_with_state: strategy = 0, decoding with 1 decoders, temperature = 0.00


whisper_full_with_state: prompt[0] = [_SOT_]
whisper_full_with_state: prompt[1] = [_LANG_en]
whisper_full_with_state: prompt[2] = [_TRANSCRIBE_]


whisper_full_with_state: id =   0, decoder = 0, token =  50364, p =  0.722, ts =    [_BEG_],  0.722, result_len =    0 '[_BEG_]'

[tazz4843]

There is no way to disable all output. Upstream whisper.cpp offers very limited control over this. Your best best is capturing all output with one of the logging hooks included in the crate and disabling whisper.cpp logging entirely. Even then, if you're using one of the specialized compute backends, you will still get some logging printed to stdout/stderr.

[tazz4843]

Going to go ahead and close this as an upstream bug, feel free to reopen if you need more help.

[vlovich]

@tazz4843 could you reconsider? I think ggml_log_set is all that needs to be called to intercept all log output & redirect it somewhere. utilityai/llama-cpp-rs#637 is an example of how to do this (llama.cpp has it's own additional logs but I believe whisper.cpp logs directly to ggml).

[tazz4843]

Oh my bad I thought we already did call this fn but clearly not. I'll take a stab at implementing it in a bit.

[tazz4843]

See #200