Please make v2 compatible with v1 （langchain use v1, while fastapi use v2）

[mokeyish]

Initial Checks

• I confirm that I'm using Pydantic V2

Description

Please make v2 compatible with v1 （langchain use v1, while fastapi use v2）. they can't work together.

Example Code

No response

Python, Pydantic & OS Version

pydantic version: 2.4.2
        pydantic-core version: 2.10.1
          pydantic-core build: profile=release pgo=true
                 install path: /home/www/.conda/envs/tmpenv/lib/python3.9/site-packages/pydantic
               python version: 3.9.18 (main, Sep 11 2023, 13:41:44)  [GCC 11.2.0]
                     platform: Linux-5.10.0-26-amd64-x86_64-with-glibc2.31
             related packages: typing_extensions-4.8.0 fastapi-0.103.1

[Luca-Blight]

Seems like the onus is on langchain.

[sydney-runkle]

Hi @mokeyish,

Thanks for submitting this issue. I agree with @Luca-Blight, this is simply a case of conflicting dependencies. You could use an earlier version of fastapi that uses pydantic V1 until langchain upgrades to supporting v2.

Any fastapi versions before 0.100.0 should use V1.

Read more here: tiangolo/fastapi#9709

[mokeyish]

I found that it also not compatiable with deepspeed. Each library has its own dependencies, and many of them now conflict with each other.

[sydney-runkle]

@mokeyish,

You can use different versions of each library to make sure the versions are compatible. The burden of resolving version conflicts isn't on pydantic, here.

[mokeyish]

Currently, fastapi/lmdeploy use v2, langchain/deepspeed use v1.

Don’t learn from Google, learn from Facebook. As I known, Google has two incompatible upgraded products, angular(v1->v2) and tensorflow(v1->v2). Facebook's react and pytorch have always been very compatible, even from pytorch v1 to v2.

pydantic is not an app and you can freely choose dependent versions. pydantic is a library that is used by many downstream libraries, some use v2, and some are still using v1. Incompatible upgrades will cause many problems.

[mokeyish]

I'm not the only one having this problem.

tiangolo/fastapi#10360

langchain-ai/langserve#10

#7901

[samuelcolvin]

Hi @mokeyish, the rewrite in Pydantic from V1 to V2 was a fundamental restructure of the codebase, virtually everything (except the unit test, the philosophy and the majority of the public API) changed.

There won't be another breaking change to Pydantic like this.

I'd love Pydantic V1 to be compatible with Pydantic V2, the problem is that doing a thorough job of it would take me and someone else 6 months of solid work. That's 6 months we don't spend improving V2. On balance I don't think that's a good use of my time.

As @Luca-Blight said, the onus has to be on dependent libraries to upgrade.

I'm not the only one having this problem.

Pydantic is currently downloaded 123m times a month. That's 47 times a second.

I'm not sure what point you were trying to make by including links to 3 issues, but the way I read it, it shows what a small proportion of Pydantic's use is affective by v1 vs. v2 compatibility.