is it possible to pass Pydantic

is it possible to pass Pydantic BaseModel as a tool input? it seems it is not supported as per the error

Plain Text

    raise RuntimeError(f'no validator found for {type_}, see `arbitrary_types_allowed` in Config')
RuntimeError: no validator found for <clas

16 comments

3345ishaan

@kapa.ai how to pass Pydantic BaseModel as tool input?

kkapa.ai

Let me check the knowledge base for you. One moment! :books:

kkapa.ai

----------------------
To pass a Pydantic BaseModel as tool input in LlamaIndex, you can use the FunctionTool class from llama_index.core.tools. Here's a step-by-step guide:

Define your Pydantic BaseModel. For example, a Song model:

Plain Text

from pydantic import BaseModel

class Song(BaseModel):
    """A song with name and artist"""
    name: str
    artist: str

Define a function that takes the fields of your BaseModel as arguments and returns an instance of your BaseModel. For example, a generate_song function:

Plain Text

def generate_song(name: str, artist: str) -> Song:
    """Generates a song with provided name and artist."""
    return Song(name=name, artist=artist)

Create a FunctionTool from your function using the from_defaults method:

Plain Text

from llama_index.core.tools import FunctionTool

tool = FunctionTool.from_defaults(fn=generate_song)

Use the predict_and_call method of your LLM (either Upstage or OpenAI) to call the tool with your input:

Plain Text

from llama_index.llms.upstage import Upstage

llm = Upstage()
response = llm.predict_and_call([tool], "Generate a song")

In this example, "Generate a song" is the input to the LLM, which will be passed to the generate_song function. The function will return a Song object, which is an instance of the Pydantic BaseModel.

You can find more details in the LlamaIndex documentation () and GitHub examples ().
----------------------
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
----------------------

LLogan M

I don't see why not. Might need to see the full error though

3345ishaan

This is the error:

Plain Text

    raise RuntimeError(f'no validator found for {type_}, see `arbitrary_types_allowed` in Config')
RuntimeError: no validator found for <class 'external_data_loader.GithubRepoItem'>, see `arbitrary_types_allowed` in Config

3345ishaan

and function tool is specified as follows:

Plain Text

def chat(self, repoItem: GithubRepoItem, query: str) -> AgentResponse:

3345ishaan

@Logan M it seems if i use the create_model from Pydantic it works, somehow seems related to the version used by llamaindex for create_model doesn't support

3345ishaan

looking at the code more, seems like it is about pydantic.v1 which errors out.

3345ishaan

Plain Text

try:
    import pydantic.v1 as pydantic
    from pydantic.v1 import (
        BaseConfig,
        BaseModel,
        Field,
        PrivateAttr,
        StrictFloat,
        StrictInt,
        StrictStr,
        create_model,
        root_validator,
        validator,
        parse_obj_as,
    )
    from pydantic.v1.error_wrappers import ValidationError
    from pydantic.v1.fields import FieldInfo
    from pydantic.v1.generics import GenericModel
except ImportError:
    import pydantic  # type: ignore
    from pydantic import (
        BaseConfig,
        BaseModel,
        Field,
        PrivateAttr,
        StrictFloat,
        StrictInt,
        StrictStr,
        create_model,
        root_validator,
        validator,
        parse_obj_as,
    )
    from pydantic.error_wrappers import ValidationError
    from pydantic.fields import FieldInfo
    from pydantic.generics import GenericModel

3345ishaan

i am not sure why pydantic.v1 gets priority over pydantic here.

3345ishaan

ok so if i use pydantic.v1 to create the BaseModel , it works. seems like a compatibility issue. i think the recommendation here should be to use from llama_index.core.bridge.pydantic import BaseModel, Field whenver constructing inputs instead of using from pydantic import BaseModel

LLogan M

Pydantic.v1 gets priority because at least right now, we support both pydantic v1 and v2 (through the compatibility layer)

LLogan M

If we don't import from .v1, a lot of things break

LLogan M

Hoping to make a full migration soon. But for now, use the v1 layer

3345ishaan

thanks. so it seems like constructing pydantic BaseModel from v2 does not work with create_model from v1.

3345ishaan

for now as i said, i am using from llama_index.core.bridge.pydantic import BaseModel, Field while defined the tool inputs in my codebase

Add a reply

Find answers from the community

is it possible to pass Pydantic