error in PydanticOutputParser

[oguzhaslak]

OutputParserException('Failed to parse Plan from completion {}. Got: 4 validation errors for Plan\nlocale\n Field required [type=missing, input_value={}, input_type=dict]\n For further information visit https://errors.pydantic.dev/2.10/v/missing\nhas_enough_context\n Field required [type=missing, input_value={}, input_type=dict]\n For further information visit https://errors.pydantic.dev/2.10/v/missing\nthought\n Field required [type=missing, input_value={}, input_type=dict]\n For further information visit https://errors.pydantic.dev/2.10/v/missing\ntitle\n Field required [type=missing, input_value={}, input_type=dict]\n For further information visit https://errors.pydantic.dev/2.10/v/missing\nFor troubleshooting, visit: https://python.langchain.com/docs/troubleshooting/errors/OUTPUT_PARSING_FAILURE ')Traceback (most recent call last):

File "/app/.venv/lib/python3.12/site-packages/langchain_core/output_parsers/pydantic.py", line 28, in _parse_obj
return self.pydantic_object.model_validate(obj)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "/app/.venv/lib/python3.12/site-packages/pydantic/main.py", line 627, in model_validate
return cls.pydantic_validator.validate_python(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

pydantic_core._pydantic_core.ValidationError: 4 validation errors for Plan
locale
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
has_enough_context
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
thought
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
title
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing

The above exception was the direct cause of the following exception:

Traceback (most recent call last):

File "/app/.venv/lib/python3.12/site-packages/langchain_core/runnables/base.py", line 1927, in _call_with_config
context.run(

File "/app/.venv/lib/python3.12/site-packages/langchain_core/runnables/config.py", line 396, in call_func_with_variable_args
return func(input, **kwargs) # type: ignore[call-arg]
^^^^^^^^^^^^^^^^^^^^^

File "/app/.venv/lib/python3.12/site-packages/langchain_core/output_parsers/base.py", line 194, in
lambda inner_input: self.parse_result(
^^^^^^^^^^^^^^^^^^

File "/app/.venv/lib/python3.12/site-packages/langchain_core/output_parsers/pydantic.py", line 68, in parse_result
return self._parse_obj(json_object)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "/app/.venv/lib/python3.12/site-packages/langchain_core/output_parsers/pydantic.py", line 36, in _parse_obj
raise self._parser_exception(e, obj) from e

langchain_core.exceptions.OutputParserException: Failed to parse Plan from completion {}. Got: 4 validation errors for Plan
locale
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
has_enough_context
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
thought
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
title
Field required [type=missing, input_value={}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.10/v/missing
For troubleshooting, visit: https://python.langchain.com/docs/troubleshooting/errors/OUTPUT_PARSING_FAILURE

[ethanwheatthin]

Same issue here using tavily as the web search engine

[MagicCube]

From the screenshot you provided, the Plan object in JSON format seems to be invalid, or at least missing some fields. This is usually caused by the LLM output not meeting the format instructed by the system prompt.

Could you please provide more debugging information? Like the specific model and version of the LLM you are using, and the original text output by the large model, which will help us debug further.

[oguzhaslak]

From the screenshot you provided, the Plan object in JSON format seems to be invalid, or at least missing some fields. This is usually caused by the LLM output not meeting the format instructed by the system prompt.

Could you please provide more debugging information? Like the specific model and version of the LLM you are using, and the original text output by the large model, which will help us debug further.

.env file

Application Settings

DEBUG=True
APP_ENV=development

docker build args

NEXT_PUBLIC_API_URL="http://192.168.1.51:8000/api"

Search Engine, Supported values: tavily (recommended), duckduckgo, brave_search, arxiv

SEARCH_API=duckduckgo
#TAVILY_API_KEY=tvly-xxx

BRAVE_SEARCH_API_KEY=xxx # Required only if SEARCH_API is brave_search

JINA_API_KEY=jina_xxx # Optional, default is None

Optional, volcengine TTS for generating podcast

#VOLCENGINE_TTS_APPID=xxx
#VOLCENGINE_TTS_ACCESS_TOKEN=xxx

VOLCENGINE_TTS_CLUSTER=volcano_tts # Optional, default is volcano_tts

VOLCENGINE_TTS_VOICE_TYPE=BV700_V2_streaming # Optional, default is BV700_V2_streaming

Option, for langsmith tracing and monitoring

LANGSMITH_TRACING=true
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
LANGSMITH_API_KEY="xxxxxxxxxxxxxxxxxxxxxxx"
LANGSMITH_PROJECT="default"

config.yaml file

BASIC_MODEL:
model: "qwen3:14b"
base_url: "http://192.168.1.51:11434/v1" # Local service address of Ollama, which can be started/viewed via ollama serve
api_key: "xxxxxxxxxxxxxxxxxxxxxx"

langchain log.

https://smith.langchain.com/public/c156580d-e2e7-464c-ac77-778c6a98351b/r

coordinator

runnablesequance

[ethanwheatthin]

I added some logging to the planner_node function and it seems messages that are being passed from the background investigation may be causing the parser to fail when trying to parse the plan generated from the planner prompt.

2025-05-15 10:01:24,137 - src.graph.nodes - INFO - STEP 1: Initializing planner node
2025-05-15 10:01:24,138 - src.graph.nodes - INFO - STEP 2: Processing iteration 0 of 1 maximum
2025-05-15 10:01:24,142 - src.graph.nodes - INFO - STEP 3: Processing background investigation if available
2025-05-15 10:01:24,143 - src.graph.nodes - INFO - Added background investigation results to messages
2025-05-15 10:01:24,143 - src.graph.nodes - INFO - STEP 4: Initializing LLM (type: basic)
2025-05-15 10:01:24,143 - src.graph.nodes - INFO - STEP 5: Invoking LLM with 3 messages
2025-05-15 10:01:44,565 - httpx - INFO - HTTP Request: POST http://localhost:11434/v1/chat/completions "HTTP/1.1 200 OK"
2025-05-15 10:01:44,587 - src.graph.nodes - ERROR - FAILED STEP 5: LLM invocation failed: Failed to parse Plan from completion {"@context": "alternativeMedia", "@type": "AlternativeContent", "@name": "Speed of Light vs Sound", "@url": null, "datePublished": null, "image": [{"@type": "ImageObject", "url": "https://taylorsciencegeeks.weebly.com/uploads/5/9/2/0/59201005/252941363.jpg"}, {"@type": "ImageObject", "url": null, "image_description": null}, {"@type": "ImageObject", "url": "https://image1.slideserve.com/2386506/speed-light-vs-sound1-l.jpg"}, {"@type": "ImageObject", "url": "http://taylorsciencegeeks.weebly.com/uploads/5/9/2/0/59201005/211520427.gif"}, {"@type": "ImageObject", "url": "https://image.slidesharecdn.com/light-lesson1-100110092314-phpapp01/95/light-lesson-1-4-728.jpg?cb=1263115616"}, {"@type": "ImageObject", "url": "https://image3.slideserve.com/5427815/speed-of-sound-vs-speed-of-light-l.jpg"}, "]}", {"@context": "AlternativeContent", "@type": "Answer", "@name": "Speed of Light vs Sound - Amazing Visuals and Speed Comparison", "datePublished": "2023-01-23T16:04:48+00:00", "image": [{"@type": "ImageObject", "url": "https://taylorsciencegeeks.weebly.com/uploads/5/9/2 /0/59201005/252941363.jpg"}], "author": null, "mainEntity": [{"@type": "Question", "@name": "How fast does sound travel?", "acceptanceMessage": {"@type": "DefinedTerm", "term": "light speed compared to sound speed", "definition": "Light travels much faster than sound, which is a necessary fact you have not known about."}, "url": "https://www.taylor-science-geeks.com/light-lesson1.html"}, "answer", {"@type": "Answer", "@context": null, "mainEntityOfPage": {"@type": "WebPage", "id": "Aidens-Teaching-Materials-Like-the-Visualizations-Here", "url": "https://www.taylor-science-geeks.com/light-lesson1.html"}, "mainContent": "Light travels much faster than sound,"}], "url": null, "mainEntityOfPage": {"@type": "FAQPage", "id": null, "publisher": {"@type": "Organization", "name": "AIDEN"}, "encodingFormat": null, "description": null, "alternateName": null, "url": null, "sameAs": [null, "{'@type': 'Answer', '@context': 'https://schema.org/QuestionAnswer', 'question': 'How fast does sound travel?'}", null], "mainContent": null, "headerImageContent": {"@type": "MediaObject", "name": "Light speed comparison", "imageLocation": null, "contentUrl": null}, "mainEntity": [{"@type": "QuestionAnswer", "question": [{"@type": "Question", "name": null, "acceptanceMessage": {"@type": "DefinedTerm", "term": null, "definition": null}, "url": null}, "answer", {"@type": "Answer", "@": null, "mainEntityOfPage": {"@type": "Webpage", "id": "Light Speeds Are Very Very Different From Each Other 2020 ", "name": "Amazing Visualizations of the Incredible Difference Between the Speed of "}}]}]}}]}. Got: 4 validation errors for Plan
locale
Field required [type=missing, input_value={'@context': 'alternative... the Speed of '}}]}]}}]}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.11/v/missing
has_enough_context
Field required [type=missing, input_value={'@context': 'alternative... the Speed of '}}]}]}}]}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.11/v/missing
thought
Field required [type=missing, input_value={'@context': 'alternative... the Speed of '}}]}]}}]}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.11/v/missing
title
Field required [type=missing, input_value={'@context': 'alternative... the Speed of '}}]}]}}]}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.11/v/missing
For troubleshooting, visit: https://python.langchain.com/docs/troubleshooting/errors/OUTPUT_PARSING_FAILURE

[Vy-Bui17]

Hey @MagicCube @ethanwheatthin , any updates on this? I have got the same issue unfotunately.

[0xlws2]

#99

i think its the same issue, in my case the error was because the llm didnt respond in the requested json format

• had to simplify the system prompt

[taxman20000]

Fixed!

I had the same issue and solved the malformed json output by changing the planner's method from json_mode to json_schema and optionally adding strict=True in this section of the code:

deer-flow/src/graph/nodes.py

Lines 105 to 108 in 7d38e5f

	llm = get_llm_by_type(AGENT_LLM_MAP["planner"]).with_structured_output(
	Plan,
	method="json_mode",
	)

Fix:

The following corrected version works with the Qwen3-235B model on the provider venice.ai:

        llm = get_llm_by_type(AGENT_LLM_MAP["planner"]).with_structured_output(
            Plan,
            method="json_schema", # change the method to json_schema
            strict=True, # optional: add this line if json_schema alone does not fix the issue, but this option might not be supported by all inference providers
        )

Background:

I am not sure that strict=True is actually necessary. According to the langchain documentation that parameter is necessary to force the sampler to choose tokens which are consistent with the provided schema, but in my testing the change to method="json_schema" was sufficient to resolve the malformed json outputs.

This fix should work for all models (even very small models) because the schema is enforced by the sampler that chooses the tokens. The schema is NOT enforced by the model itself. That means the efficacy of this fix depends on the implementation of the sampler (i.e. your inference provider) and not on the model you use.

@Vy-Bui17 @ethanwheatthin @MagicCube

[taxman20000]

PR has been merged so the fix is live. The model validation error should no longer occur.

I have submitted a PR that fixes the pydantic validation error. Waiting for review by the maintainers: #322

[taxman20000]

Previous PR #322 had to be reverted. I have since submitted a new improved fix in PR #341