Prompt splitting multiple delimiters

[SlayerTO]

had an occurence of multiple ### in prompt, rsplit would be one solution. No solution is perfect because there is no way of finding out which is the real delimiter.

[SlayerTO]

soome clients do not add the whitespace before and after the delimteter. would introduce another unwanted bahavior

[MaikoTan]

Hello, thanks for the PR.

So you meant that some prompts would contain multiple ###s which causes part of the "positive" prompt would become the negative prompt? Is there any other problem with this solution? Could you let me know why closed the PR?

[SlayerTO]

For sure @MaikoTan

I tried this solution first but very quickly I realized that you can get valid proompts where there are no white spaces for example "cat###dog" usually the prompt will be delivered like this "cat ### dogs" so that solution will fail in the first case, because i don't know how to edit an already oopend pull request i closed it.

I'm playing around modified it to run multi model and with one request I had the problem that someone did manage to write a prompt like this "white###cat ### dog" that did fail obvios because the map got 3 instead of 2 parms.

A solution would be

prompt, negative = map(lambda x: x.strip(), prompt.rsplit(###", 1))

however we can't with certainty say if the additioonal delimiter is in the positiv or negativ part. But with an rsplit or an split limited to 1 we could at least run the request.

The optimal solution would be if we could flag the request as an maleformed prompt for the API but i didn't saw anything like this.

Now im unsure if i should open another patch for the one liner, if you want that at all or if you implement a solution yourself.

best regards