Overview
Output parsers for Perplexity LLM responses that strip reasoning <think> tags before delegating to standard JSON and Pydantic parsers.
Description
This module, part of the langchain-perplexity partner package, provides specialized output parsers that handle Perplexity model responses containing chain-of-thought reasoning enclosed in <think>...</think> tags. It includes a helper function strip_think_tags and two parser classes: ReasoningJsonOutputParser (extends JsonOutputParser) and ReasoningStructuredOutputParser (extends PydanticOutputParser). Both parsers strip reasoning tags and markdown code fences before passing the cleaned text to their parent parsers.
Usage
Import these parsers when working with Perplexity models that produce chain-of-thought reasoning in <think> blocks and you need to extract structured JSON or Pydantic-validated output from the response.
Code Reference
Source Location
- Repository: Langchain_ai_Langchain
- File:
libs/partners/perplexity/langchain_perplexity/output_parsers.py
- Lines: 1-88
Signature
def strip_think_tags(text: str) -> str:
"""Removes all <think>...</think> tags and their content from text."""
...
class ReasoningJsonOutputParser(JsonOutputParser):
"""A JSON output parser that strips reasoning tags before parsing."""
def parse_result(self, result: list[Generation], *, partial: bool = False) -> Any:
...
class ReasoningStructuredOutputParser(PydanticOutputParser[TBaseModel], Generic[TBaseModel]):
"""A structured output parser that strips reasoning tags before parsing."""
def parse_result(self, result: list[Generation], *, partial: bool = False) -> Any:
...
Import
from langchain_perplexity.output_parsers import (
strip_think_tags,
ReasoningJsonOutputParser,
ReasoningStructuredOutputParser,
)
I/O Contract
strip_think_tags
Inputs
| Name |
Type |
Required |
Description
|
| text |
str |
Yes |
The input text that may contain <think>...</think> tags and markdown code fences.
|
Outputs
| Name |
Type |
Description
|
| return |
str |
The text with all <think>...</think> blocks and markdown code fences removed.
|
ReasoningJsonOutputParser.parse_result
Inputs
| Name |
Type |
Required |
Description
|
| result |
list[Generation] |
Yes |
The result of the LLM call containing generated text.
|
| partial |
bool |
No |
Whether to parse partial JSON objects. Defaults to False.
|
Outputs
| Name |
Type |
Description
|
| return |
Any |
The parsed JSON object with reasoning tags stripped.
|
ReasoningStructuredOutputParser.parse_result
Inputs
| Name |
Type |
Required |
Description
|
| result |
list[Generation] |
Yes |
The result of the LLM call containing generated text.
|
| partial |
bool |
No |
Whether to parse partial JSON objects. Defaults to False.
|
Outputs
| Name |
Type |
Description
|
| return |
Any |
The parsed Pydantic model instance with reasoning tags stripped.
|
Usage Examples
Basic Usage
from langchain_perplexity.output_parsers import (
strip_think_tags,
ReasoningJsonOutputParser,
)
# Strip think tags from raw text
raw = "<think>Let me reason about this...</think>{\"name\": \"Alice\"}"
cleaned = strip_think_tags(raw)
# cleaned == '{"name": "Alice"}'
# Use the parser in a chain
parser = ReasoningJsonOutputParser()
# parser can be used as part of a LangChain chain:
# chain = llm | parser
Structured Output with Pydantic
from pydantic import BaseModel
from langchain_perplexity.output_parsers import ReasoningStructuredOutputParser
class UserInfo(BaseModel):
name: str
age: int
parser = ReasoningStructuredOutputParser(pydantic_object=UserInfo)
# Use in a chain with a Perplexity model that outputs reasoning tags
# chain = llm | parser
Related Pages
Page Connections
Double-click a node to navigate. Hold to expand connections.
