Remove printing of prompt and prompt tokenization at startup

[slaren]

Now that the tokenizer has been tested fairly well, printing the tokenization on startup adds a lot of clutter for no good reason.

Additionally also removes printing of the prompt itself since it is already printed as it is evaluated anyway.

[ggerganov]

The lines have to be either commented so we can easily re-enable - sometime it is still useful to look at the tokens.
I think it is best to make this on/off via command line arg

[anzz1]

Do not agree, a '--quiet' option would be better instead. It's very good information to know which tokens are generated when trying out different prompts and such. The tokens are vital in researching the whole thing.

Actually it would be a cool addition to add a cmdline option which would show the output tokenized along their id's to get a better understanding of different models and their differences. So another option is to go the other way around, instead of '--quiet' it would be a '-v' / '--verbose' option to show this info.

Also as the "debug" and "standard" outputs are already directed to different streams, you know that you can easily show only the generated output by redirecting the stderr output to a file or nul?

file:
main -m ./models/llama-13B-ggml/ggml-model-q4_0.bin 2> err.log

silent:
main -m ./models/llama-13B-ggml/ggml-model-q4_0.bin 2>nul (windows)
main -m ./models/llama-13B-ggml/ggml-model-q4_0.bin 2>/dev/null (unix)

In any case I agree that it definitely shouldn't be outright removed but made a command line option. But it's already easy to redirect the stderr to elsewhere if you want a 'clean look' when not testing/researching.

[ggerganov]

Added a command line option for now: 502a400