core[minor]: support pydantic v2 models in PydanticOutputParser #18811

jnis23 · 2024-03-08T18:11:18Z

As mentioned in #18322, the current PydanticOutputParser won't work for anyone trying to parse to pydantic v2 models. This PR adds a separate PydanticV2OutputParser, as well as a langchain_core.pydantic_v2 namespace that will fail on import to any projects using pydantic<2. Happy to update the docs for output parsers if this is something we're interesting in adding.

On a separate note, I also updated check_pydantic.sh to detect pydantic imports with leading whitespace and excluded the internal namespaces. That change can be separated into its own PR if needed.

vercel · 2024-03-08T18:11:22Z

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Ignored Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Mar 27, 2024 2:13pm

eyurtsev

Will require investigation

eyurtsev · 2024-03-08T18:34:06Z

libs/core/langchain_core/output_parsers/pydantic_v2.py

@@ -0,0 +1,55 @@
+import json


A bit worried about getting this into the code base. Definitely into core as it will create duplicated code in core when it upgrades to pydantic 2

Not sure, but this could potentially create issues when using in a chain or a runnable map together since it'll cause mixing of pydantic v1 and v2 code.

eyurtsev · 2024-03-08T18:36:26Z

@jnis23 Apologies for blocking. I think we need to do a bit of investigation first to figure out the pathway for a pydantic v2 migration and how this would interact with it.

jnis23 · 2024-03-08T19:39:27Z

@eyurtsev No problem at all. Agreed it feels bad to dupe that much code, but couldn't think of a better way without adding some additional abstraction. Lmk if there is anything I can do to help re: the investigation!

eyurtsev · 2024-03-08T19:44:41Z

@jnis23 we basically have two options:

spec out full support for pydantic v2 for LangChain (at least in terms of roadmap)
get some partial fixes in and check that they don't fail in unexpected ways.

If you want to investigate (2), I'd check that it still works with something like prompt | model | parser chain and write a unit test for that. I'm mostly worried about type inference (via OutputType/InputType) ending up and mixing pydnatic v1 with pydantic v2 types (which would crash things), I don't know if it can happen, but would love some code to double check it.

In this case, could we not do something like:

TBaseModel = TypeVar("TBaseModel", bound=PydanticBaseModel)

with PydanticBaseModel being Union[BaseModelv2, BaseModelv1] if pydantic version == 2, and
PydanticBaseModel = BaseModelv1 if pydantic version == 1?

other conditional behavior based on version, so it's still a single interface

eyurtsev · 2024-03-08T19:45:34Z

for

spec out full support for pydantic v2 for LangChain (at least in terms of roadmap)

I'll try to check a potential pathway forward for this next week, and will share if it looks promising

jnis23 · 2024-03-08T20:16:18Z

@jnis23 we basically have two options:
1. spec out full support for pydantic v2 for LangChain (at least in terms of roadmap)

2. get some partial fixes in and check that they don't fail in unexpected ways.
If you want to investigate (2), I'd check that it still works with something like prompt | model | parser chain and write a unit test for that. I'm mostly worried about type inference (via OutputType/InputType) ending up and mixing pydnatic v1 with pydantic v2 types (which would crash things), I don't know if it can happen, but would love some code to double check it.

I'm currently using the v2 parser in a chain without issue so I can add a test for that. Are there supported use cases for outputs getting piped back into other runnables? I can definitely see that being a problem if the runnable depends on v1. If you know of any example off the top of your head, I can start there, otherwise I can dig.

jnis23 · 2024-03-11T19:35:04Z

@eyurtsev Was able to get both versions working under the existing PydanticOutputParser. Pyright can figure out the typing, but Mypy gets hung up on the type alias being re-declared. I also don't love that we need the conditional checks in _parse_obj for both the BaseModel and the ValidationError, but it should get the job done until a full transition to V2.

If this looks good, I can nuke the V2 class

eyurtsev · 2024-03-14T02:05:37Z

libs/core/langchain_core/pydantic_v2/__init__.py

+# * This change is easier to roll out and roll back.
+
+
+if get_pydantic_major_version() < 2:


There should be no side effects on import since the import is triggered by default rather than user action

eyurtsev · 2024-03-14T02:05:59Z

libs/core/langchain_core/pydantic_v2/__init__.py

@@ -0,0 +1,24 @@
+import warnings


I'm not convinced that the namespace is needed yet -- could you review the code snippets -- i think some form of either those versions or else local scope imports will work

eyurtsev · 2024-03-14T02:10:07Z

libs/core/langchain_core/output_parsers/pydantic.py

+        if PYDANTIC_MAJOR_VERSION == 2 and issubclass(
+            self.pydantic_object, V2BaseModel
+        ):
+            try:


The validation error can be imported here locally -- the import is cached so there's no issue in terms of speed

if VERSION = 2: if issubclass(self.pydantic_object, V1): ..... except pydantic.v1.ValidationError: else: except pydantic.ValidationError:

eyurtsev · 2024-03-14T02:13:16Z

libs/core/langchain_core/output_parsers/pydantic.py


 from langchain_core.exceptions import OutputParserException
 from langchain_core.output_parsers import JsonOutputParser
 from langchain_core.outputs import Generation
 from langchain_core.pydantic_v1 import BaseModel, ValidationError
+from langchain_core.pydantic_v2 import BaseModel as V2BaseModel
+from langchain_core.pydantic_v2 import ValidationError as V2ValidationError
+from langchain_core.utils.pydantic import PYDANTIC_MAJOR_VERSION



pydantic_v2 module could probably be eliminated to do something like this instead:

import pydantic if VERSION = 2: PydanticBaseModel = Union[pydantic.BaseModel, pydantic.v1.BaseModel] else: PydanticBaseModel = pydantic.BaseModel

@eyurtsev Definitely like this approach better, it's much cleaner. The namespace was to get around the check_pydantic script that gets run as part of linting and in CI. If we're fine adding an exclusion for this file and the test, I can push up these changes

Yeah OK skipping this file or adding some # pydantic: ignore type filter

okay, added support for # pydantic: ignore. Lmk how the rest looks

eyurtsev · 2024-03-14T02:16:31Z

@jnis23 looks good. you can remove the older version of the code. I added a few suggestions that I think can further reduce the code foot print of the change, let me know if those work

jnis23 · 2024-03-27T14:36:37Z

@eyurtsev ready for you to re-review when you can

eyurtsev · 2024-03-27T19:38:07Z

Thank you @jnis23!

…chain-ai#18811) As mentioned in langchain-ai#18322, the current PydanticOutputParser won't work for anyone trying to parse to pydantic v2 models. This PR adds a separate `PydanticV2OutputParser`, as well as a `langchain_core.pydantic_v2` namespace that will fail on import to any projects using pydantic<2. Happy to update the docs for output parsers if this is something we're interesting in adding. On a separate note, I also updated `check_pydantic.sh` to detect pydantic imports with leading whitespace and excluded the internal namespaces. That change can be separated into its own PR if needed. --------- Co-authored-by: Jan Nissen <[email protected]>

As mentioned in #18322, the current PydanticOutputParser won't work for anyone trying to parse to pydantic v2 models. This PR adds a separate `PydanticV2OutputParser`, as well as a `langchain_core.pydantic_v2` namespace that will fail on import to any projects using pydantic<2. Happy to update the docs for output parsers if this is something we're interesting in adding. On a separate note, I also updated `check_pydantic.sh` to detect pydantic imports with leading whitespace and excluded the internal namespaces. That change can be separated into its own PR if needed. --------- Co-authored-by: Jan Nissen <[email protected]>

eyurtsev · 2024-06-07T22:17:47Z

Hi @jnis23 not sure if you have any cycles to look into this: #22669

Potentially related to this PR

Jan Nissen added 3 commits March 8, 2024 10:52

add v2 parser

b1af8b5

check whitespace, ignore namespace/utils

20df86e

add pydantic_v2 namespace and output parser

c2b3211

dosubot bot added size:M This PR changes 30-99 lines, ignoring generated files. Ɑ: parsing Related to output parser module 🤖:improvement Medium size change to existing code to handle new use-cases labels Mar 8, 2024

hard fail -> warning, reformat message

90e25ea

eyurtsev requested changes Mar 8, 2024

View reviewed changes

eyurtsev added investigate Flagged for investigation. Ɑ: pydantic Related to pydantic or pydantic migration labels Mar 8, 2024

eyurtsev self-assigned this Mar 8, 2024

update parser name, add test for v2 parser chaining

d870dc2

dosubot bot added size:L This PR changes 100-499 lines, ignoring generated files. and removed size:M This PR changes 30-99 lines, ignoring generated files. labels Mar 8, 2024

Jan Nissen added 3 commits March 8, 2024 18:15

test fmt

c7ea1b6

merge pydantic v1 and v2 parsers to one interface

4fdc8dc

| -> Union

de98790

eyurtsev reviewed Mar 14, 2024

View reviewed changes

Jan Nissen and others added 4 commits March 20, 2024 12:38

add pydantic: ignore

748392a

rm pydantic_v2, unify v1 & v2 output parser

6bff30a

Merge branch 'master' into pydantic-v2-parser

de798dc

fix imports after merge

7121c93

simplify validationerror catch

7981d34

jnis23 changed the title ~~core: add PydanticV2OutputParser~~ core: support pydantic v2 models in PydanticOutputParser Mar 26, 2024

remove unneccesary import

2174c54

eyurtsev self-requested a review March 27, 2024 19:34

eyurtsev approved these changes Mar 27, 2024

View reviewed changes

dosubot bot added the lgtm PR looks good. Use to confirm that a PR is ready for merging. label Mar 27, 2024

eyurtsev changed the title ~~core: support pydantic v2 models in PydanticOutputParser~~ core[minor]: support pydantic v2 models in PydanticOutputParser Mar 27, 2024

eyurtsev merged commit 2e0ddd6 into langchain-ai:master Mar 27, 2024
95 checks passed

jnis23 deleted the pydantic-v2-parser branch March 27, 2024 20:19

rpgoldman mentioned this pull request Apr 19, 2024

New mypy type error from PydanticOutputParser #20634

Closed

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

core[minor]: support pydantic v2 models in PydanticOutputParser #18811

core[minor]: support pydantic v2 models in PydanticOutputParser #18811

jnis23 commented Mar 8, 2024

vercel bot commented Mar 8, 2024 •

edited

Loading

eyurtsev left a comment

eyurtsev Mar 8, 2024

eyurtsev commented Mar 8, 2024

jnis23 commented Mar 8, 2024

eyurtsev commented Mar 8, 2024 •

edited

Loading

eyurtsev commented Mar 8, 2024

jnis23 commented Mar 8, 2024

jnis23 commented Mar 11, 2024

eyurtsev Mar 14, 2024

eyurtsev Mar 14, 2024

eyurtsev Mar 14, 2024

eyurtsev Mar 14, 2024

jnis23 Mar 14, 2024

eyurtsev Mar 19, 2024

jnis23 Mar 20, 2024

eyurtsev commented Mar 14, 2024

jnis23 commented Mar 27, 2024

eyurtsev commented Mar 27, 2024

eyurtsev commented Jun 7, 2024

		# * This change is easier to roll out and roll back.


		if get_pydantic_major_version() < 2:

core[minor]: support pydantic v2 models in PydanticOutputParser #18811

core[minor]: support pydantic v2 models in PydanticOutputParser #18811

Conversation

jnis23 commented Mar 8, 2024

vercel bot commented Mar 8, 2024 • edited Loading

eyurtsev left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

eyurtsev commented Mar 8, 2024

jnis23 commented Mar 8, 2024

eyurtsev commented Mar 8, 2024 • edited Loading

eyurtsev commented Mar 8, 2024

jnis23 commented Mar 8, 2024

jnis23 commented Mar 11, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

eyurtsev commented Mar 14, 2024

jnis23 commented Mar 27, 2024

eyurtsev commented Mar 27, 2024

eyurtsev commented Jun 7, 2024

vercel bot commented Mar 8, 2024 •

edited

Loading

eyurtsev commented Mar 8, 2024 •

edited

Loading