Pydantic compatibility: No Breaking Changes

[baskaryan]

[UPDATE 08/20/23]

As of 0.0.267 LangChain does not pin to pydantic v1 or v2, and users should be able to install and use either. Compatibility with pydantic v1 and v2 will be maintained until at least the end of 2023. We have delayed all immediate plans to migrate to pydantic v2, and will provide further updates if/when we have them.

We've heard from a number of users and partners that due to other dependencies they are unable to migrate their own projects to pydantic v2 at the moment. We deeply appreciate the feedback, and will continue to do our best to support projects on both pydantic v1 and v2.

---

[ORIGINAL POST]

• Pydantic v2 was released in June 2023 (https://docs.pydantic.dev/2.0/blog/pydantic-v2-final/)
• v2 has a number of breaking changes (https://docs.pydantic.dev/2.0/migration/)
• Pydantic v2 and v1 are under the same package name, so both versions cannot be installed at the same time

LangChain Pydantic Migration Plan

Langchain will carry out the migration to pydantic v2 in two steps:

1. 2023-08-17: LangChain will allow users to install either Pydantic V1 or V2 as of version 0.0.267
   • Internally LangChain will continue to use V1.
   • During this time, users can pin their pydantic version to v1 to avoid breaking changes, or start a partial
     migration using pydantic v2 throughout their code, but avoiding mixing v1 and v2 code for LangChain (see below).
2. 2023-08-25: Langchain will migrate internally to using V2 code.
   • Users will have to upgrade to pydantic>1 as well to use LangChain. This does not mean that all user code must use v2 semantics.
   • Users should stop using pydantic.v1 namespace interacting with LangChain.
   • However, users can KEEP on using any of their current v1 code using pydantic.v1 namespace! (see link below)
   • This will allow users to upgrade to pydantic v2 semantics for custom code at their own pace. The following package: bump-pydantic package to help with the upgrade process; however, the upgrade is far from being fully automated for large code base.

Migration Plan Suggested for Users

If your code base is large, or you are responsible for bumping many libraries at once, consider migrating in 2 steps:

1. Upgrade to pydantic v2, but keep on using v1 namespace. Here's the way we did it in LangChain. This allows all the code to remain more or less the same while allowing for installation of pydantic>=2
2. When ready, deprecate usage of v1 namespace, and transition the code to actually use pydantic v2 fucntionality.

Between 2023-08-17 and 2023-08-25 releases

User can either pin to pydantic v1, and upgrade their code in one go once LangChain has migrated to v2 internally, or they can start a partial migration to v2, but must avoid mixing v1 and v2 code for LangChain.

Below are two examples of showing how to avoid mixing pydantic v1 and v2 code in
the case of inheritance and in the case of passing objects to LangChain.

Example 1: Extending via inheritance

YES

from pydantic.v1 import root_validator, validator

class CustomTool(BaseTool): # BaseTool is v1 code
    x: int = Field(default=1)

    def _run(*args, **kwargs):
        return "hello"

    @validator('x') # v1 code
    @classmethod
    def validate_x(cls, x: int) -> int:
        return 1

CustomTool(
    name='custom_tool',
    description="hello",
    x=1,
)

Mixing Pydantic v2 primitives with Pydantic v1 primitives can raise cryptic errors

NO

from pydantic import Field, field_validator # pydantic v2

class CustomTool(BaseTool): # BaseTool is v1 code
    x: int = Field(default=1)

    def _run(*args, **kwargs):
        return "hello"

    @field_validator('x') # v2 code
    @classmethod
    def validate_x(cls, x: int) -> int:
        return 1

CustomTool(
    name='custom_tool',
    description="hello",
    x=1,
)

Example 2: Passing objects to LangChain

YES

from langchain.tools.base import Tool
from pydantic.v1 import BaseModel, Field # <-- Uses v1 namespace

class CalculatorInput(BaseModel):
    question: str = Field()

Tool.from_function( # <-- tool uses v1 namespace
    func=lambda question: 'hello',
    name="Calculator",
    description="useful for when you need to answer questions about math",
    args_schema=CalculatorInput
)

NO

from langchain.tools.base import Tool
from pydantic import BaseModel, Field # <-- Uses v2 namespace

class CalculatorInput(BaseModel):
    question: str = Field()

Tool.from_function( # <-- tool uses v1 namespace
    func=lambda question: 'hello',
    name="Calculator",
    description="useful for when you need to answer questions about math",
    args_schema=CalculatorInput
)

After 2023-08-25 release

• Users must upgrade to v2
• Users should not pass pydantic.v1 derived objects to LangChain or rely on pydantic.v1 when extending functionality

[jscheel]

Am I correct in understanding that to stay up to date, all our application code must be upgraded to pydantic 2 within 1.5 weeks? I understand langchain is 0.0.x, but that seems a bit much, yeah? Is there a plan for locking in versioning a bit more after this?

[shroominic]

I think its possible to migrate really fast using this

[srhinos]

this is a pretty crazy timetable for migrations to pydantic v2.

this is coming from someone who put together the original migration of langchain to v2 a while back here - #8330

theres still several downstream deps that aren't upgraded to v2 and won't allow upgrades to occur.

What should be happening here is a step between 1 and 2 where while pydantic is unlocked to allow upgrades to v2 but isn't running v2, downstream dependancies that are blocking the full migration to v2 are identified and patches pushed where necessary. As noted on my PR above, there are still several repos that need these support patches. ChromaDB being the most awful as they had similar downstream upgrades required.

Only once this new step completes should the timer to force upgrades to v2 be started.

[baskaryan]

@srhinos thanks for the feedback!

As of langchain==0.0.267 the pydantic version is uncapped and we should largely be cross-compatible with pydantic v1 and v2 (modulo the known failure modes mentioned above). Would love to hear if that works for you. We're open to delaying the full migration if the current solution works sufficiently well and enough integrations / dependents aren't able to migrate to v2 yet.

[jscheel]

Just and additional data point, haystack is not upgrading to pydantic 2 until they do a full major version change deepset-ai/haystack#5304. Guessing that this is gonna be a mess for people.

[baskaryan]

Please note the updated discussion text:

[UPDATE 08/20/23]

As of 0.0.267 LangChain does not pin to pydantic v1 or v2, and users should be able to install and use either. Compatibility with pydantic v1 and v2 will be maintained until at least the end of 2023. We have delayed all immediate plans to migrate to pydantic v2, and will provide further updates if/when we have them.

We've heard from a number of users and partners that due to other dependencies they are unable to migrate their own projects to pydantic v2 at the moment. We deeply appreciate the feedback, and will continue to do our best to support projects on both pydantic v1 and v2.

[baskaryan]

cc @srhinos @jscheel

[jscheel]

@baskaryan thanks for the update. I appreciate your willingness to reconsider this, and I appreciate that you've added compatibility to work unpinned. 🚀 🚀 🚀

[simjak]

Hey community,
How to solve this pydantic stuff:

  File ".venv/run.py", line 6, in <module>
    from langchain.chat_models import ChatOpenAI
  File ".venv/lib/python3.11/site-packages/langchain/__init__.py", line 6, in <module>
    from langchain.agents import MRKLChain, ReActChain, SelfAskWithSearchChain
  File ".venv/lib/python3.11/site-packages/langchain/agents/__init__.py", line 31, in <module>
    from langchain.agents.agent import (
  File ".venv/lib/python3.11/site-packages/langchain/agents/agent.py", line 14, in <module>
    from langchain.agents.agent_iterator import AgentExecutorIterator
  File ".venv/lib/python3.11/site-packages/langchain/agents/agent_iterator.py", line 21, in <module>
    from langchain.callbacks.manager import (
  File ".venv/lib/python3.11/site-packages/langchain/callbacks/__init__.py", line 10, in <module>
    from langchain.callbacks.aim_callback import AimCallbackHandler
  File ".venv/lib/python3.11/site-packages/langchain/callbacks/aim_callback.py", line 5, in <module>
    from langchain.schema import AgentAction, AgentFinish, LLMResult
  File ".venv/lib/python3.11/site-packages/langchain/schema/__init__.py", line 5, in <module>
    from langchain.schema.memory import BaseChatMessageHistory, BaseMemory
  File ".venv/lib/python3.11/site-packages/langchain/schema/memory.py", line 7, in <module>
    from langchain.schema.messages import AIMessage, BaseMessage, HumanMessage
  File ".venv/lib/python3.11/site-packages/langchain/schema/messages.py", line 148, in <module>
    class HumanMessageChunk(HumanMessage, BaseMessageChunk):
  File ".venv/lib/python3.11/site-packages/pydantic/main.py", line 352, in __new__
    cls = super().__new__(mcs, name, bases, new_namespace, **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen abc>", line 106, in __new__
TypeError: multiple bases have instance lay-out conflict```

[nikokaoja]

I am migrating my project to use pydantic v2, and have bump into the issue of creating MyChat class which inherits BaseChatModel and adding additional validators. I've reported issue here.

Any help how to overcome this is more than appreciated.

[zhuhaow]

Will langchain ever be upgraded to use pydantic v2?

Given langchain only works with pydantic 1. There is currently no way we can progressively migrate our code which depending on langchain from pydantic v1 to v2. So there won't be too much change no matter how long we wait.

If we do want to migrate to pydantic v2, we may need an active timeline.

[eyurtsev]

We're collecting feedback here #24405 to determine which issues users are facing and determine how and whether to prioritize the migration.

Could you explain why LangChain prevents upgrading to pydantic 2?

[ariofrio]

FWIW, I'm trying to share models I wrote for FastAPI endpoints in LangChain tools, but FastAPI requires Pydantic v2 and LangChain requires using Pydantic v1 (or pydantic.v1). It's been 10 months since the original planned date for upgrading LangChain to Pydantic v2 internally. Any movement or updates on when this migration, originally planned for 2023-08-25, will happen?

From what I can tell, there's two issues here:

1. The actual work of migrating Langchain to Pydantic 2
2. The decision on when to "flip the switch", which will require downstream users to upgrade to Pydantic 2 as well.

[eyurtsev]

We're collecting feedback here #24405

Could you give an example of how you're exposing your tools as API end points? Do you need the arg schema only or the entire tool as a pydantic model?