Using pydantic>2 requires specifying fastapi version, too

[nickstenning]

I encountered today the surprising fact that the following cog.yaml results in a broken image:

build:
  gpu: false
  python_version: "3.12"
  python_packages:
    - "pydantic>2"
predict: "predict.py:Predictor"

The build fails with the following error:

[...]
#14 DONE 0.7s
Validating model schema...

Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/usr/local/lib/python3.12/site-packages/cog/command/openapi_schema.py", line 13, in <module>
    from ..server.http import create_app
  File "/usr/local/lib/python3.12/site-packages/cog/server/http.py", line 18, in <module>
    from fastapi import Body, FastAPI, Header, Path, Response
  File "/usr/local/lib/python3.12/site-packages/fastapi/__init__.py", line 7, in <module>
    from .applications import FastAPI as FastAPI
  File "/usr/local/lib/python3.12/site-packages/fastapi/applications.py", line 16, in <module>
    from fastapi import routing
  File "/usr/local/lib/python3.12/site-packages/fastapi/routing.py", line 22, in <module>
    from fastapi import params
  File "/usr/local/lib/python3.12/site-packages/fastapi/params.py", line 4, in <module>
    from pydantic.fields import FieldInfo, Undefined
ImportError: cannot import name 'Undefined' from 'pydantic.fields' (/usr/local/lib/python3.12/site-packages/pydantic/fields.py)

ⅹ Failed to get type signature: exit status 1

As far as I can tell the problem here is that fastapi==0.98 is installed by default in the cog-base images, but this is actually not compatible with Pydantic v2.

It's possible to make a working image by specifying a tighter range of acceptable versions for fastapi, but I don't think this is ideal for users:

build:
  gpu: false
  python_version: "3.12"
  python_packages:
    - "fastapi>0.100.0,<0.111.0"
    - "pydantic>2"
predict: "predict.py:Predictor"

[8W9aG]

#2115