BaseLLM bug about langflow HOT 7 CLOSED

I fixed the tooltip part, but changing type={'str'} part will trigger some side effects within the handling code so I decided not to touch that until I'm certain about its behavior 😄

Will take a further investigation afterwards if needed :)

from langflow.

Hey!
We'll look into that ASAP.

@ibiscp do you know what could be causing this? I tested it here and it happened to me too.

I ran pip install langflow -Uand langflow put the OpenAI node and the PAL-MATH node and I can't connect them to each other.

Maybe the output to OpenAI shouldn't be str.

from langflow.

In frontend/src/CustomNodes/GenericNode.js L84

id={data.type + "|" + data.id + "|" + data.node.base_classes.map((b) => ("|" + b))}

This line is problematic because it produces id in the form like this OpenAI|dndnode_26|BaseOpenAI,|BaseLLM,|BaseLanguageModel
You can see the trailing commas making isValidConnection to return false-negatives on valid connections.
So changing that line to

id={data.type + "|" + data.id + data.node.base_classes.join('|')}

or more likely to

id={[data.type, data.id, ...data.node.base_classes].join('|')}

will fix the problem in current implementation.

We also need to fix hard-coded "type:str" notation on ParameterComponent. I'd like to also find a way to actually use type parameter rather than just parse handle's id field when we're validating connections between nodes (Because current implementation is eager to be broken in future if we try to introduce way to manage custom components. Using reserved eparators for component ID will break the validation), but I'm not sure that's doable as this is my first time using reactflow library.

May I open PR from my fork to resolve this issue or is someone already working on it?

from langflow.

I'll debug that.

from langflow.

Thanks for the help! I'll test it out now.
Let's take the hard-coded "type:str" discussion to another issue or is it related to this?

from langflow.

Ok. I'll open a discussion based on your comment so we can maybe think of a better solution.

from langflow.

Fixed in version 0.0.44

from langflow.