Value not declarable with JSON Schema (Pydantic V2)

[JP-Ellis]

Summary

I am unable to get LangServe to work nicely with a Pydantic V2 model.

Background

I understand that LangChain and LangServe are undergoing an upgrade from Pydantic V1 to V2. I was under the impression that LangServe supports V2 models, with the one caveat that OpenAPI docs aren't presently generated.

Example

Here is a simple reproducible example of the error, along with all installed dependencies and the error logs. The TL;DR is:

• Dependencies:
  • langchain-core==0.2.9
  • langserve==0.2.2
  • pydantic==2.7.4
• Error:

  raise ValueError(f'Value not declarable with JSON Schema, field: {field}')
  ValueError: Value not declarable with JSON Schema, field: name='__root__' type=Optional[ExampleInput] required=False default=None
  

Virtual Environment

❯ which python
/Users/JP-Ellis/tmp/.venv/bin/python

❯ python --version
Python 3.12.4

❯ uv pip install fastapi langchain_core langserve pydantic sse_starlette
Resolved 55 packages in 36ms
Installed 55 packages in 44ms
 + annotated-types==0.7.0
 + anyio==4.4.0
 + attrs==23.2.0
 + certifi==2024.6.2
 + charset-normalizer==3.3.2
 + click==8.1.7
 + dnspython==2.6.1
 + email-validator==2.1.2
 + fastapi==0.111.0
 + fastapi-cli==0.0.4
 + h11==0.14.0
 + httpcore==1.0.5
 + httptools==0.6.1
 + httpx==0.27.0
 + idna==3.7
 + jinja2==3.1.4
 + jsonpatch==1.33
 + jsonpointer==3.0.0
 + jsonschema==4.22.0
 + jsonschema-specifications==2023.12.1
 + langchain-core==0.2.9
 + langserve==0.2.2
 + langsmith==0.1.81
 + markdown-it-py==3.0.0
 + markupsafe==2.1.5
 + mdurl==0.1.2
 + orjson==3.10.5
 + packaging==24.1
 + pydantic==2.7.4
 + pydantic-core==2.18.4
 + pygments==2.18.0
 + pyproject-toml==0.0.10
 + python-dotenv==1.0.1
 + python-multipart==0.0.9
 + pyyaml==6.0.1
 + referencing==0.35.1
 + requests==2.32.3
 + rich==13.7.1
 + rpds-py==0.18.1
 + setuptools==70.1.0
 + shellingham==1.5.4
 + sniffio==1.3.1
 + sse-starlette==2.1.2
 + starlette==0.37.2
 + tenacity==8.4.1
 + toml==0.10.2
 + typer==0.12.3
 + typing-extensions==4.12.2
 + ujson==5.10.0
 + urllib3==2.2.2
 + uvicorn==0.30.1
 + uvloop==0.19.0
 + watchfiles==0.22.0
 + websockets==12.0
 + wheel==0.43.0

example.py

from fastapi import APIRouter
from langchain_core.runnables import Runnable, RunnableConfig
from langserve import add_routes
from pydantic import BaseModel

router = APIRouter()


class ExampleInput(BaseModel):
    foo: str


class ExampleOutput(BaseModel):
    bar: str


class ExampleGenerator(Runnable[ExampleInput, ExampleOutput]):
    def invoke(
        self, input: ExampleInput, config: RunnableConfig | None = None
    ) -> ExampleOutput:
        return ExampleOutput(bar=f"Received foo={input.foo}")


add_routes(
    router,
    ExampleGenerator(),
    path="/example",
)

Error logs

❯ python example.py
Traceback (most recent call last):
  File "/Users/JP-Ellis/tmp/example.py", line 26, in <module>
    add_routes(
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/langserve/server.py", line 443, in add_routes
    api_handler = APIHandler(
                  ^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/langserve/api_handler.py", line 667, in __init__
    input_type_ = _resolve_model(
                  ^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/langserve/api_handler.py", line 338, in _resolve_model
    hash_ = model.schema_json()
            ^^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/main.py", line 675, in schema_json
    cls.schema(by_alias=by_alias, ref_template=ref_template), default=pydantic_encoder, **dumps_kwargs
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/main.py", line 664, in schema
    s = model_schema(cls, by_alias=by_alias, ref_template=ref_template)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/schema.py", line 188, in model_schema
    m_schema, m_definitions, nested_models = model_process_schema(
                                             ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/schema.py", line 581, in model_process_schema
    m_schema, m_definitions, nested_models = model_type_schema(
                                             ^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/schema.py", line 622, in model_type_schema
    f_schema, f_definitions, f_nested_models = field_schema(
                                               ^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/schema.py", line 255, in field_schema
    f_schema, f_definitions, f_nested_models = field_type_schema(
                                               ^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/schema.py", line 527, in field_type_schema
    f_schema, f_definitions, f_nested_models = field_singleton_schema(
                                               ^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/JP-Ellis/tmp/.venv/lib/python3.12/site-packages/pydantic/v1/schema.py", line 951, in field_singleton_schema
    raise ValueError(f'Value not declarable with JSON Schema, field: {field}')
ValueError: Value not declarable with JSON Schema, field: name='__root__' type=Optional[ExampleInput] required=False default=None

[eyurtsev]

LangChain does not currently support pydantic 2 models. If you're using pydantic 2, you need to use the pydantic.v1 namespace

Do this in your code;

from pydantic.v1 import BaseModel

[JP-Ellis]

Is there a tracking issue for when LangServe will support Pydantic 2? Or is there a way to use the pydantic.v1 namespace in Pydantic V2 installations?

I'm asking because I have found that the most recent release of LangChain no longer support Pydantic V1 at all due to this commit which introduced the following change to the langchain-core dependency:

pydantic = [
  { version = ">=1,<3", python =  "<3.12.4" },
  { version = "^2.7.4", python = ">=3.12.4" },
]

[eyurtsev]

Use pydantic.v1 right now with pydantic 2. We first need to migrate langchain to support pydantic 2 proper (rather than from the v1 namespace).

[JP-Ellis]

Unfortunately, using the pydantic.v1 namespace does not work for me at the moment, due to some mix up between V1 and V2 functions deep in dependencies which I would rather avoid monkey-patching. Happy to raise another issue to track that specifically if you want.

[eyurtsev]

Is it coming from langchain dependencies or another dependency?

[JP-Ellis]

In my case, it comes from the underlying FastAPI dependency. I am getting the same error as reported in:

• Working with Pydantic v1 while having v2 installed fastapi/fastapi#10360

Let me see if I can generate a minimal example within LangChain's context.

EDIT: On investigating the issue, I have confirmed it is identical to the example documented above. This does raise another question though: if FastAPI doesn't support Pydantic V1 models, how is LangServe able to avoid this very issue?

Having said all this, I am happy for you to close this issue (or keep it open for visibility, until Pydantic V2 is fully supported?)

[codekiln]

Just now saw this issue, which is very similar to the one I just filed - #725. In my case, the only reason I upgraded was that I need to update langchain-core, which was holding back many of my other lang* libraries. In order to do that, I needed to update pydantic to 2.x. I tried converting all models to from langchain.pydantic_v1 import ..., but then I discovered FastAPI was incompatible with having 2.x installed and using 1.x models; specifically, I was getting fastapi basemodel.validate() takes 2 positional arguments but 3 were given in my middleware on every request when I tried that.

[eyurtsev]

Hi this is a documented limitation currently. Please hold off until LangChain 0.3.x. If you want to use pydantic proper 2. You can use pydantic.v1 namespace but FastAPI does not support it.

https://github.com/langchain-ai/langserve?tab=readme-ov-file#pydantic
https://github.com/langchain-ai/langserve?tab=readme-ov-file#limitations
https://python.langchain.com/v0.2/docs/how_to/pydantic_compatibility/#4-langserve-cannot-generate-openapi-docs-if-running-pydantic-2

[Abd-elr4hman]

how can one help with moving langchain to pydantic v2 ? can you point me towards a good starting point to get started ?
@eyurtsev