Output Transformer

Modules concerning the output transformers used in Gen AI applications.

EventFilterOutputTransformer(event_types)

Bases: IdentityOutputTransformer

An output transformer that filters stream events by type.

This transformer lets users define the base event types to keep (e.g. thinking). At runtime, only events matching the selected types are kept, including their corresponding _start and _end events.

Attributes:

Name	Type	Description
allowed_event_types	set[str]	The set of allowed event types.

Initializes a new EventFilterOutputTransformer.

Parameters:

Name	Type	Description	Default
event_types	list[str]	Base event types to keep.	required

Raises:

Type	Description
ValueError	If any event type is empty.

IdentityOutputTransformer(event_emitter=None)

Bases: BaseOutputTransformer

An output transformer that transforms the output of a language model into an identity function.

This transformer simply returns the output and stream events of the language model as is.

JSONOutputTransformer(event_emitter=None)

Bases: IdentityOutputTransformer

An output transformer that parses the JSON string in the language model output into a Python dictionary.

This output transformer is useful to parse a structured output from the output of a language model that doesn't support structured output natively. It only supports parsing a single JSON object from the text output. Text outputs that contain multiple JSON objects will be returned as the original text output.

This output transformer will not perform any transformation for streaming events.

Attributes:

Name	Type	Description
event_emitter	EventEmitter | None	The event emitter to use for streaming events.

Examples:

LMOutput transformation: Input: python LMOutput(outputs=[ LMOutputItem(type="text", output='Here is the output: {"key": "value"}'), ])

Output:
```python
LMOutput(outputs=[
    LMOutputItem(type="structured", output={"key": "value"}),
])
```

OutputTransformerChain(transformers, event_emitter=None)

A wrapper to chain a list of output transformers.

The OutputTransformerChain class defines the interface for transforming the output of language models using a chain of output transformers.

Attributes:

Name	Type	Description
transformers	list[BaseOutputTransformer]	The list of output transformers to use.
event_emitter	EventEmitter | None	The event emitter to use for streaming events.

Initializes a new instance of the OutputTransformerChain class.

Parameters:

Name	Type	Description	Default
transformers	list[BaseOutputTransformer]	The list of output transformers to use.	required
event_emitter	EventEmitter | None	The event emitter to use for streaming events. Defaults to None.	None

emit(event) async

Transforms a stream event of a language model and emits it using the event emitter.

This method transforms a stream event of a language model using the chain of output transformers and emits it using the event emitter.

Parameters:

Name	Type	Description	Default
event	Event	The stream event to transform and emit.	required

Raises:

Type	Description
ValueError	If the event emitter is not set.

from_configs(configs, event_emitter=None) classmethod

Create an output transformer chain from a list of output transformer configs.

Parameters:

Name	Type	Description	Default
configs	list[OutputTransformerConfig]	The list of output transformer configs to use.	required
event_emitter	EventEmitter | None	The event emitter to use for streaming events. Defaults to None.	None

Returns:

Name	Type	Description
OutputTransformerChain	OutputTransformerChain	The output transformer chain.

transform(output)

Transforms the output of a language model.

This method transforms the output using the chain of output transformers.

Parameters:

Name	Type	Description	Default
output	LMOutput	The output to transform.	required

Returns:

Name	Type	Description
LMOutput	LMOutput	The transformed output of a language model.

OutputTransformerConfig

Bases: BaseModel

Defines the output transformer config schema.

Attributes:

Name	Type	Description
type	OutputTransformerType	The type of output transformer to use.
kwargs	dict[str, Any]	The additional keyword arguments for the output transformer.

build()

Build an output transformer from the config.

Returns:

Name	Type	Description
BaseOutputTransformer	BaseOutputTransformer	The output transformer.

Raises:

Type	Description
ValueError	If the output transformer type is invalid.

event_filter(event_types) classmethod

Create an event filter output transformer config.

Parameters:

Name	Type	Description	Default
event_types	list[str]	Base event types to keep.	required

Returns:

Name	Type	Description
OutputTransformerConfig	'OutputTransformerConfig'	The event filter output transformer config.

from_config(config) classmethod

Create an output transformer config from a string or a dictionary.

Parameters:

Name	Type	Description	Default
config	OutputTransformer	The output transformer config, can be one of the following: 1. str: The type of the output transformer 2. dict[str, Any]: The output transformer config as a dictionary 3. OutputTransformerConfig: The output transformer config itself	required

Returns:

Name	Type	Description
OutputTransformerConfig	'OutputTransformerConfig'	A new output transformer config.

Raises:

Type	Description
ValueError

identity() classmethod

Create an identity output transformer config.

Returns:

Name	Type	Description
OutputTransformerConfig	'OutputTransformerConfig'	The identity output transformer config.

json() classmethod

Create a JSON output transformer config.

Returns:

Name	Type	Description
OutputTransformerConfig	'OutputTransformerConfig'	The JSON output transformer config.

think_tag() classmethod

Create a think tag output transformer config.

Returns:

Name	Type	Description
OutputTransformerConfig	'OutputTransformerConfig'	The think tag output transformer config.

ThinkTagOutputTransformer(event_emitter=None)

Bases: BaseOutputTransformer

An output transformer that handles special thinking tags used in certain open source language models.

Some open source language models, such as Qwen3-30B-A3B, emits their thinking tokens as part of the text tokens, These thinking tokens are wrapped in special XML-like thinking tags to indicate the start and end of thinking. This transformer detects and separates these thinking tags from the text output to handle the thinking properly.

Attributes:

Name	Type	Description
event_emitter	EventEmitter | None	The event emitter to use for streaming events.
thinking	bool	Whether the transformer is currently in thinking mode.
response_id	str	The ID of the current response.
thinking_id	str	The ID of the current thinking.

Examples:

LMOutput transformation: Input: python LMOutput(outputs=[ LMOutputItem(type="text", output="<think>I'm thinking...</think>I'm responding..."), ])

Output:
```python
LMOutput(outputs=[
    LMOutputItem(type="thinking", output=Thinking(thinking="I'm thinking...")),
    LMOutputItem(type="text", output="I'm responding..."),
])
```

Streaming event transformation: Input: python Event(id="original_id", type="response", value="<think>") Event(id="original_id", type="response", value="I'm thinking...") Event(id="original_id", type="response", value="leading text</think>trailing text") Event(id="original_id", type="response", value="I'm responding...")

Output:
```python
Event(id="thinking_id", type="thinking_start", value="")
Event(id="thinking_id", type="thinking", value="I'm thinking...")
Event(id="thinking_id", type="thinking", value="leading text")
Event(id="thinking_id", type="thinking_end", value="")
Event(id="response_id", type="response_start", value="")
Event(id="response_id", type="response", value="trailing text")
Event(id="response_id", type="response", value="I'm responding...")
Event(id="response_id", type="response_end", value="")
```

Initializes a new instance of the ThinkTagOutputTransformer class.

Parameters:

Name	Type	Description	Default
event_emitter	EventEmitter | None	The event emitter to use for streaming events. Defaults to None.	None