Added Detection against self responses

[DVSProductions]

The models like writing "### Instruction:" from time to time. with this modification their generation gets interrupted as soon as they type ### effectively stopping them from generating their own prompts. Because this kills the llama process this also saves resources and isn't just a visual change.

Here is an example response before this change:

Here is the same prompt and the (disabled) debug output from docker showing the changes in action:

[voarsh2]

If https://github.com/nsarrazin/serge/pull/143 is merged in, then your generate.py changes won't be valid

[gaby]

@DVSProductions Can you fix the merge conflicts?

[DVSProductions]

I've resolved the merge conflicts but the Docker push seems to fail with a 403? don't know that to do about that

[gaby]

@DVSProductions Seems like another bug, for some reason it fails with external branches

[gaby]

@DVSProductions I will merge this soon, trying to fix a last little bug with the CI :-)

[fishscene]

@gaby How are things going?

[gaby]

@fishscene This branch has changes that are not compatible with the migration to Redis.

It's up to @DVSProductions to fix the merge conflicts

[fishscene]

Thanks for the update, mate!

[gaby]

@DVSProductions Do you want to fix the conflicts or close this?