[feat(json_adapter): Enhance parsing for single-field outputs & Pydantic v1/v2 compat #8239
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 2 commits
May 18, 2025 18:22
…ic v1/v2 compatibility
This commit introduces several improvements to the JSONAdapter:
1. **Improved Parsing for Single Output Fields:**
The `parse()` method now correctly handles scenarios where a Language Model
(LM) returns a raw JSON array or a primitive type (e.g., string, number,
boolean) when the DSPy Signature expects only a single output field.
Previously, this could lead to parsing errors if the LM didn't wrap the
single output in a JSON object. This change improves compatibility with
models like Gemini, which may exhibit such behavior.
2. **Pydantic v1/v2 Compatibility for Structured Outputs:**
The `_get_structured_outputs_response_format()` method has been updated
to dynamically use Pydantic's `ConfigDict` for v2.x and the older
`BaseConfig` for v1.x. This ensures that the structured output model
generation (used with OpenAI's function calling/tool use features)
is compatible across different Pydantic versions by correctly setting
`extra="forbid"`.
3. **Enhanced Error Logging:**
Improved error logging in the `__call__` method by adding
`traceback.format_exc(e)` when the initial attempt to use structured
outputs fails. This provides more detailed diagnostic information for
debugging.
- dspy/adapters/json_adapter.py: remove `logger.debug(traceback.format_exc(e))` to avoid issue with test lacking stack trace. - dspy/adapters/json_adapter.py: tighten the conditional to `if isinstance(fields, list) and len(signature.output_fields) == 1` so only list results get wrapped under a single output field, preventing improper handling of non-list completions
estsauver
changed the title
okhat
reviewed
May 22, 2025
| lm_response=completion, | ||
| message="LM response cannot be serialized to a JSON object.", | ||
| ) | ||
| if isinstance(fields, list) and len(signature.output_fields) == 1: |
There was a problem hiding this comment.
Isn't this only applicable if the (single) output field has type list ?
There was a problem hiding this comment.
Or are you saying basically let's just rely on the validation below to check anyway?
There was a problem hiding this comment.
I think so, maybe it would be better to run the check on output fields directly instead of on the "fields" variable.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
[feat(json_adapter): Enhance parsing for single-field outputs & Pydantic v1/v2 compat
This commit introduces several improvements to the JSONAdapter:
Improved Parsing for Single Output Fields:
The
parse()method now correctly handles scenarios where a Language Model(LM) returns a raw JSON array when the DSPy Signature expects only a single output field of a list type.
Previously, this could lead to parsing errors if the signature didn't wrap the
single output in a JSON object.
Pydantic v1/v2 Compatibility for Structured Outputs:
The
_get_structured_outputs_response_format()method has been updatedto dynamically use Pydantic's
ConfigDictfor v2.x and the olderBaseConfigfor v1.x. This ensures that the structured output modelgeneration (used with OpenAI's function calling/tool use features)
is compatible across different Pydantic versions by correctly setting
extra="forbid".