Input failure loop fix #298
Conversation
… mode switching later.
…ency in formatting.
|
Okay great! We're currently setting up a benchmark system for testing the performance of changes like this, so we ensure we're always moving in the right direction. |
Torantulino
added
the
Needs Benchmark
|
dont worry about it for now, still not really in the best spot |
…livers mixed message format.
There was a problem hiding this comment.
Overall the scope is just too big. But removing .gitignore change is usually a must for it to merge. Submit your .gitignore separately. And resolve the conflicts. Thanks to the big scope the conflicts are expected to be terrible.
It says it just fixes the loop, but it adds a dozen of unrelated things, such as prompt templating and whatnot. Split.
.gitignore
Outdated
There was a problem hiding this comment.
Is this an unrelated change? Rollback
|
|
||
| Returns: | ||
|
|
||
| str: AI's response. |
There was a problem hiding this comment.
Why the alignment? Doesn't """" have provisions for such messages?
github-actions
bot
added
the
conflicts
|
This pull request has conflicts with the base branch, please resolve those so we can evaluate the pull request. |
1 similar comment
|
This pull request has conflicts with the base branch, please resolve those so we can evaluate the pull request. |
|
This is a mass message from the AutoGPT core team. For more details (and for infor on joining our Discord), please refer to: |
|
@jmagly we're currently building challenges in order to measure the effect of these types of changes, and I would like to discuss with you. Please join use on Discord through this link https://discord.gg/autogpt (if not already) DM me on the Auto-GPT discord channel (my discord is merwanehamadi). |
|
my 2c: I would suggest to team up with other folks who are working on overlapping efforts and come up with a common infrastructure to easily have several prompt configs side by side, as per: #1874 (comment) |
|
This is an interesting PR, but unfortunately it has fallen out of date and the merge conflicts have become too complex to fix. You're welcome to join us on Discord if you are still interested to contribute! |
So there are a few issues here that going into a larger design discussion so for now I re-worked the prompts, added a bit more option there and created a fallback output if the LLM decides to break format.
This also removes the LLM cycle where it tries to fix its own output. Def would love to play with that again for sure. The hardcoded "you are naughty dont break format again" should do the trick for now.
Learned something doing this one, you cant give it an ounce of an idea a human is at the console, it just overrides the world. You cant even use words like "human", breaks it right out of the box.