main : add --conversation / -cnv flag

[dawidpotocki]

Closes #7100

This flag enables interactive mode and disables printing special tokens and suffix/prefix.
The name for the flag is not great, but couldn't think of anything good.

People in the issue seemed to be interested in something like this.

Example output:

> Hello!
Hello! It's nice to meet you. Is there something I can help you with, or would you like to chat?

> Nah, I'm good.
No worries then! It's always okay to just chill and say hi.

>

Potentially special token printing should also be disabled for --chatml and --instruct.

[Jeximo]

This is a nice feature. Reduces the clutter when chatting with Llama3.

[mirek190]

thanks

[olinorwell]

I feel like this flag should not enable interactive mode. That already has it's own flag.

All it should do is what the original request wanted it to do, which was effectively "disables printing special tokens and suffix/prefix".

The interactive mode being enabled is a result of the use case it fell into, not the requested feature.

Appreciate the effort but I believe it should be called something like 'no special token output' and do what it was originally doing, which was only outputting generated tokens that were not considered 'special', thus no <|eot_id|> at the end for example.

The flag could then be used in any of the modes as and when it was required.

[dawidpotocki]

Welp, that's fair.

I guess such flag could still be added as I don't think changing the
behaviour of this one would make much sense.