ConstitutionalChain

Note

ConstitutionalChain implements the standard Runnable Interface. 🏃

The Runnable Interface has additional methods that are available on runnables, such as with_types, with_retry, assign, bind, get_graph, and more.

class langchain.chains.constitutional_ai.base.ConstitutionalChain

Bases: Chain

Chain for applying constitutional principles.

Example

from langchain_community.llms import OpenAI
from langchain.chains import LLMChain, ConstitutionalChain
from langchain.chains.constitutional_ai.models                 import ConstitutionalPrinciple

llm = OpenAI()

qa_prompt = PromptTemplate(
    template="Q: {question} A:",
    input_variables=["question"],
)
qa_chain = LLMChain(llm=llm, prompt=qa_prompt)

constitutional_chain = ConstitutionalChain.from_llm(
    llm=llm,
    chain=qa_chain,
    constitutional_principles=[
        ConstitutionalPrinciple(
            critique_request="Tell if this answer is good.",
            revision_request="Give a better answer.",
        )
    ],
)

constitutional_chain.run(question="What is the meaning of life?")

param callback_manager: BaseCallbackManager | None = None

[DEPRECATED] Use callbacks instead.

param callbacks: Callbacks = None

Optional list of callback handlers (or callback manager). Defaults to None. Callback handlers are called throughout the lifecycle of a call to a chain, starting with on_chain_start, ending with on_chain_end or on_chain_error. Each custom chain can optionally call additional callback methods, see Callback docs for full details.

param chain: LLMChain [Required]

param constitutional_principles: List[ConstitutionalPrinciple] [Required]

param critique_chain: LLMChain [Required]

param memory: BaseMemory | None = None

Optional memory object. Defaults to None. Memory is a class that gets called at the start and at the end of every chain. At the start, memory loads variables and passes them along in the chain. At the end, it saves any returned variables. There are many different types of memory - please see memory docs for the full catalog.

param metadata: Dict[str, Any] | None = None

Optional metadata associated with the chain. Defaults to None. This metadata will be associated with each call to this chain, and passed as arguments to the handlers defined in callbacks. You can use these to eg identify a specific instance of a chain with its use case.

param return_intermediate_steps: bool = False

param revision_chain: LLMChain [Required]

param tags: List[str] | None = None

Optional list of tags associated with the chain. Defaults to None. These tags will be associated with each call to this chain, and passed as arguments to the handlers defined in callbacks. You can use these to eg identify a specific instance of a chain with its use case.

param verbose: bool [Optional]

Whether or not run in verbose mode. In verbose mode, some intermediate logs will be printed to the console. Defaults to the global verbose value, accessible via langchain.globals.get_verbose().

__call__(inputs: Dict[str, Any] | Any, return_only_outputs: bool = False, callbacks: List[BaseCallbackHandler] | BaseCallbackManager | None = None, *, tags: List[str] | None = None, metadata: Dict[str, Any] | None = None, run_name: str | None = None, include_run_info: bool = False) → Dict[str, Any]

Deprecated since version langchain==0.1.0: Use invoke instead.

Execute the chain.

Parameters:

• inputs (Dict[str, Any] | Any) – Dictionary of inputs, or single input if chain expects only one param. Should contain all inputs specified in Chain.input_keys except for inputs that will be set by the chain’s memory.

• return_only_outputs (bool) – Whether to return only outputs in the response. If True, only new keys generated by this chain will be returned. If False, both input keys and new keys generated by this chain will be returned. Defaults to False.

• callbacks (List[BaseCallbackHandler] | BaseCallbackManager | None) – Callbacks to use for this chain run. These will be called in addition to callbacks passed to the chain during construction, but only these runtime callbacks will propagate to calls to other objects.

• tags (List[str] | None) – List of string tags to pass to all callbacks. These will be passed in addition to tags passed to the chain during construction, but only these runtime tags will propagate to calls to other objects.

• metadata (Dict[str, Any] | None) – Optional metadata associated with the chain. Defaults to None

• include_run_info (bool) – Whether to include run info in the response. Defaults to False.

• run_name (str | None)

Returns:

A dict of named outputs. Should contain all outputs specified in

Chain.output_keys.

Return type:

Dict[str, Any]

async abatch(inputs: List[Input], config: RunnableConfig | List[RunnableConfig] | None = None, *, return_exceptions: bool = False, **kwargs: Any | None) → List[Output]

Default implementation runs ainvoke in parallel using asyncio.gather.

The default implementation of batch works well for IO bound runnables.

Subclasses should override this method if they can batch more efficiently; e.g., if the underlying Runnable uses an API which supports a batch mode.

Parameters:

• inputs (List[Input]) – A list of inputs to the Runnable.

• config (RunnableConfig | List[RunnableConfig] | None) – A config to use when invoking the Runnable. The config supports standard keys like ‘tags’, ‘metadata’ for tracing purposes, ‘max_concurrency’ for controlling how much work to do in parallel, and other keys. Please refer to the RunnableConfig for more details. Defaults to None.

• return_exceptions (bool) – Whether to return exceptions instead of raising them. Defaults to False.

• kwargs (Any | None) – Additional keyword arguments to pass to the Runnable.

Returns:

A list of outputs from the Runnable.

Return type:

List[Output]

async abatch_as_completed(inputs: Sequence[Input], config: RunnableConfig | Sequence[RunnableConfig] | None = None, *, return_exceptions: bool = False, **kwargs: Any | None) → AsyncIterator[Tuple[int, Output | Exception]]

Run ainvoke in parallel on a list of inputs, yielding results as they complete.

Parameters:

• inputs (Sequence[Input]) – A list of inputs to the Runnable.

• config (RunnableConfig | Sequence[RunnableConfig] | None) – A config to use when invoking the Runnable. The config supports standard keys like ‘tags’, ‘metadata’ for tracing purposes, ‘max_concurrency’ for controlling how much work to do in parallel, and other keys. Please refer to the RunnableConfig for more details. Defaults to None. Defaults to None.

• return_exceptions (bool) – Whether to return exceptions instead of raising them. Defaults to False.

• kwargs (Any | None) – Additional keyword arguments to pass to the Runnable.

Yields:

A tuple of the index of the input and the output from the Runnable.

Return type:

AsyncIterator[Tuple[int, Output | Exception]]

async acall(inputs: Dict[str, Any] | Any, return_only_outputs: bool = False, callbacks: List[BaseCallbackHandler] | BaseCallbackManager | None = None, *, tags: List[str] | None = None, metadata: Dict[str, Any] | None = None, run_name: str | None = None, include_run_info: bool = False) → Dict[str, Any]

Deprecated since version langchain==0.1.0: Use ainvoke instead.

Asynchronously execute the chain.

Parameters:

• inputs (Dict[str, Any] | Any) – Dictionary of inputs, or single input if chain expects only one param. Should contain all inputs specified in Chain.input_keys except for inputs that will be set by the chain’s memory.

• return_only_outputs (bool) – Whether to return only outputs in the response. If True, only new keys generated by this chain will be returned. If False, both input keys and new keys generated by this chain will be returned. Defaults to False.

• callbacks (List[BaseCallbackHandler] | BaseCallbackManager | None) – Callbacks to use for this chain run. These will be called in addition to callbacks passed to the chain during construction, but only these runtime callbacks will propagate to calls to other objects.

• tags (List[str] | None) – List of string tags to pass to all callbacks. These will be passed in addition to tags passed to the chain during construction, but only these runtime tags will propagate to calls to other objects.

• metadata (Dict[str, Any] | None) – Optional metadata associated with the chain. Defaults to None

• include_run_info (bool) – Whether to include run info in the response. Defaults to False.

• run_name (str | None)

Returns:

A dict of named outputs. Should contain all outputs specified in

Chain.output_keys.

Return type:

Dict[str, Any]

async ainvoke(input: Dict[str, Any], config: RunnableConfig | None = None, **kwargs: Any) → Dict[str, Any]

Default implementation of ainvoke, calls invoke from a thread.

The default implementation allows usage of async code even if the Runnable did not implement a native async version of invoke.

Subclasses should override this method if they can run asynchronously.

Parameters:

• input (Dict[str, Any])

• config (RunnableConfig | None)

• kwargs (Any)

Return type:

Dict[str, Any]

apply(input_list: List[Dict[str, Any]], callbacks: List[BaseCallbackHandler] | BaseCallbackManager | None = None) → List[Dict[str, str]]

Deprecated since version langchain==0.1.0: Use batch instead.

Call the chain on all inputs in the list.

Parameters:

• input_list (List[Dict[str, Any]])

• callbacks (List[BaseCallbackHandler] | BaseCallbackManager | None)

Return type:

List[Dict[str, str]]

async aprep_inputs(inputs: Dict[str, Any] | Any) → Dict[str, str]

Prepare chain inputs, including adding inputs from memory.

Parameters:

inputs (Dict[str, Any] | Any) – Dictionary of raw inputs, or single input if chain expects only one param. Should contain all inputs specified in Chain.input_keys except for inputs that will be set by the chain’s memory.

Returns:

A dictionary of all inputs, including those added by the chain’s memory.

Return type:

Dict[str, str]

async aprep_outputs(inputs: Dict[str, str], outputs: Dict[str, str], return_only_outputs: bool = False) → Dict[str, str]

Validate and prepare chain outputs, and save info about this run to memory.

Parameters:

• inputs (Dict[str, str]) – Dictionary of chain inputs, including any inputs added by chain memory.

• outputs (Dict[str, str]) – Dictionary of initial chain outputs.

• return_only_outputs (bool) – Whether to only return the chain outputs. If False, inputs are also added to the final outputs.

Returns:

A dict of the final chain outputs.

Return type:

Dict[str, str]

async arun(*args: Any, callbacks: List[BaseCallbackHandler] | BaseCallbackManager | None = None, tags: List[str] | None = None, metadata: Dict[str, Any] | None = None, **kwargs: Any) → Any

Deprecated since version langchain==0.1.0: Use ainvoke instead.

Convenience method for executing chain.

The main difference between this method and Chain.__call__ is that this method expects inputs to be passed directly in as positional arguments or keyword arguments, whereas Chain.__call__ expects a single input dictionary with all the inputs

Parameters:

• *args (Any) – If the chain expects a single input, it can be passed in as the sole positional argument.

• callbacks (List[BaseCallbackHandler] | BaseCallbackManager | None) – Callbacks to use for this chain run. These will be called in addition to callbacks passed to the chain during construction, but only these runtime callbacks will propagate to calls to other objects.

• tags (List[str] | None) – List of string tags to pass to all callbacks. These will be passed in addition to tags passed to the chain during construction, but only these runtime tags will propagate to calls to other objects.

• **kwargs (Any) – If the chain expects multiple inputs, they can be passed in directly as keyword arguments.

• metadata (Dict[str, Any] | None)

• **kwargs

Returns:

The chain output.

Return type:

Any

Example

# Suppose we have a single-input chain that takes a 'question' string:
await chain.arun("What's the temperature in Boise, Idaho?")
# -> "The temperature in Boise is..."

# Suppose we have a multi-input chain that takes a 'question' string
# and 'context' string:
question = "What's the temperature in Boise, Idaho?"
context = "Weather report for Boise, Idaho on 07/03/23..."
await chain.arun(question=question, context=context)
# -> "The temperature in Boise is..."

as_tool(args_schema: Type[BaseModel] | None = None, *, name: str | None = None, description: str | None = None, arg_types: Dict[str, Type] | None = None) → BaseTool

Beta

This API is in beta and may change in the future.

Create a BaseTool from a Runnable.

as_tool will instantiate a BaseTool with a name, description, and args_schema from a Runnable. Where possible, schemas are inferred from runnable.get_input_schema. Alternatively (e.g., if the Runnable takes a dict as input and the specific dict keys are not typed), the schema can be specified directly with args_schema. You can also pass arg_types to just specify the required arguments and their types.

Parameters:

• args_schema (Optional[Type[BaseModel]]) – The schema for the tool. Defaults to None.

• name (Optional[str]) – The name of the tool. Defaults to None.

• description (Optional[str]) – The description of the tool. Defaults to None.

• arg_types (Optional[Dict[str, Type]]) – A dictionary of argument names to types. Defaults to None.

Returns:

A BaseTool instance.

Return type:

BaseTool

Typed dict input:

from typing import List
from typing_extensions import TypedDict
from langchain_core.runnables import RunnableLambda

class Args(TypedDict):
    a: int
    b: List[int]

def f(x: Args) -> str:
    return str(x["a"] * max(x["b"]))

runnable = RunnableLambda(f)
as_tool = runnable.as_tool()
as_tool.invoke({"a": 3, "b": [1, 2]})

dict input, specifying schema via args_schema:

from typing import Any, Dict, List
from langchain_core.pydantic_v1 import BaseModel, Field
from langchain_core.runnables import RunnableLambda

def f(x: Dict[str, Any]) -> str:
    return str(x["a"] * max(x["b"]))

class FSchema(BaseModel):
    """Apply a function to an integer and list of integers."""

    a: int = Field(..., description="Integer")
    b: List[int] = Field(..., description="List of ints")

runnable = RunnableLambda(f)
as_tool = runnable.as_tool(FSchema)
as_tool.invoke({"a": 3, "b": [1, 2]})

dict input, specifying schema via arg_types:

from typing import Any, Dict, List
from langchain_core.runnables import RunnableLambda

def f(x: Dict[str, Any]) -> str:
    return str(x["a"] * max(x["b"]))

runnable = RunnableLambda(f)
as_tool = runnable.as_tool(arg_types={"a": int, "b": List[int]})
as_tool.invoke({"a": 3, "b": [1, 2]})

String input:

from langchain_core.runnables import RunnableLambda

def f(x: str) -> str:
    return x + "a"

def g(x: str) -> str:
    return x + "z"

runnable = RunnableLambda(f) | g
as_tool = runnable.as_tool()
as_tool.invoke("b")

Added in version 0.2.14.

async astream(input: Input, config: RunnableConfig | None = None, **kwargs: Any | None) → AsyncIterator[Output]

Default implementation of astream, which calls ainvoke. Subclasses should override this method if they support streaming output.

Parameters:

• input (Input) – The input to the Runnable.

• config (RunnableConfig | None) – The config to use for the Runnable. Defaults to None.

• kwargs (Any | None) – Additional keyword arguments to pass to the Runnable.

Yields:

The output of the Runnable.

Return type:

AsyncIterator[Output]

astream_events(input: Any, config: RunnableConfig | None = None, *, version: Literal['v1', 'v2'], include_names: Sequence[str] | None = None, include_types: Sequence[str] | None = None, include_tags: Sequence[str] | None = None, exclude_names: Sequence[str] | None = None, exclude_types: Sequence[str] | None = None, exclude_tags: Sequence[str] | None = None, **kwargs: Any) → AsyncIterator[StandardStreamEvent | CustomStreamEvent]

Beta

This API is in beta and may change in the future.

Generate a stream of events.

Use to create an iterator over StreamEvents that provide real-time information about the progress of the Runnable, including StreamEvents from intermediate results.

A StreamEvent is a dictionary with the following schema:

• event: str - Event names are of the

  format: on_[runnable_type]_(start|stream|end).

• name: str - The name of the Runnable that generated the event.

• run_id: str - randomly generated ID associated with the given execution of

  the Runnable that emitted the event. A child Runnable that gets invoked as part of the execution of a parent Runnable is assigned its own unique ID.

• parent_ids: List[str] - The IDs of the parent runnables that

  generated the event. The root Runnable will have an empty list. The order of the parent IDs is from the root to the immediate parent. Only available for v2 version of the API. The v1 version of the API will return an empty list.

• tags: Optional[List[str]] - The tags of the Runnable that generated

  the event.

• metadata: Optional[Dict[str, Any]] - The metadata of the Runnable

  that generated the event.

• data: Dict[str, Any]

Below is a table that illustrates some evens that might be emitted by various chains. Metadata fields have been omitted from the table for brevity. Chain definitions have been included after the table.

ATTENTION This reference table is for the V2 version of the schema.

event	name	chunk	input	output
on_chat_model_start	[model name]		{“messages”: [[SystemMessage, HumanMessage]]}
on_chat_model_stream	[model name]	AIMessageChunk(content=”hello”)
on_chat_model_end	[model name]		{“messages”: [[SystemMessage, HumanMessage]]}	AIMessageChunk(content=”hello world”)
on_llm_start	[model name]		{‘input’: ‘hello’}
on_llm_stream	[model name]	‘Hello’
on_llm_end	[model name]		‘Hello human!’
on_chain_start	format_docs
on_chain_stream	format_docs	“hello world!, goodbye world!”
on_chain_end	format_docs		[Document(…)]	“hello world!, goodbye world!”
on_tool_start	some_tool		{“x”: 1, “y”: “2”}
on_tool_end	some_tool			{“x”: 1, “y”: “2”}
on_retriever_start	[retriever name]		{“query”: “hello”}
on_retriever_end	[retriever name]		{“query”: “hello”}	[Document(…), ..]
on_prompt_start	[template_name]		{“question”: “hello”}
on_prompt_end	[template_name]		{“question”: “hello”}	ChatPromptValue(messages: [SystemMessage, …])

In addition to the standard events, users can also dispatch custom events (see example below).

Custom events will be only be surfaced with in the v2 version of the API!

A custom event has following format:

Attribute	Type	Description
name	str	A user defined name for the event.
data	Any	The data associated with the event. This can be anything, though we suggest making it JSON serializable.

Here are declarations associated with the standard events shown above:

format_docs:

def format_docs(docs: List[Document]) -> str:
    '''Format the docs.'''
    return ", ".join([doc.page_content for doc in docs])

format_docs = RunnableLambda(format_docs)

some_tool:

@tool
def some_tool(x: int, y: str) -> dict:
    '''Some_tool.'''
    return {"x": x, "y": y}

prompt:

template = ChatPromptTemplate.from_messages(
    [("system", "You are Cat Agent 007"), ("human", "{question}")]
).with_config({"run_name": "my_template", "tags": ["my_template"]})

Example:

from langchain_core.runnables import RunnableLambda

async def reverse(s: str) -> str:
    return s[::-1]

chain = RunnableLambda(func=reverse)

events = [
    event async for event in chain.astream_events("hello", version="v2")
]

# will produce the following events (run_id, and parent_ids
# has been omitted for brevity):
[
    {
        "data": {"input": "hello"},
        "event": "on_chain_start",
        "metadata": {},
        "name": "reverse",
        "tags": [],
    },
    {
        "data": {"chunk": "olleh"},
        "event": "on_chain_stream",
        "metadata": {},
        "name": "reverse",
        "tags": [],
    },
    {
        "data": {"output": "olleh"},
        "event": "on_chain_end",
        "metadata": {},
        "name": "reverse",
        "tags": [],
    },
]

Example: Dispatch Custom Event

from langchain_core.callbacks.manager import (
    adispatch_custom_event,
)
from langchain_core.runnables import RunnableLambda, RunnableConfig
import asyncio


async def slow_thing(some_input: str, config: RunnableConfig) -> str:
    """Do something that takes a long time."""
    await asyncio.sleep(1) # Placeholder for some slow operation
    await adispatch_custom_event(
        "progress_event",
        {"message": "Finished step 1 of 3"},
        config=config # Must be included for python < 3.10
    )
    await asyncio.sleep(1) # Placeholder for some slow operation
    await adispatch_custom_event(
        "progress_event",
        {"message": "Finished step 2 of 3"},
        config=config # Must be included for python < 3.10
    )
    await asyncio.sleep(1) # Placeholder for some slow operation
    return "Done"

slow_thing = RunnableLambda(slow_thing)

async for event in slow_thing.astream_events("some_input", version="v2"):
    print(event)

Parameters:

• input (Any) – The input to the Runnable.

• config (RunnableConfig | None) – The config to use for the Runnable.

• version (Literal['v1', 'v2']) – The version of the schema to use either v2 or v1. Users should use v2. v1 is for backwards compatibility and will be deprecated in 0.4.0. No default will be assigned until the API is stabilized. custom events will only be surfaced in v2.

• include_names (Sequence[str] | None) – Only include events from runnables with matching names.

• include_types (Sequence[str] | None) – Only include events from runnables with matching types.

• include_tags (Sequence[str] | None) – Only include events from runnables with matching tags.

• exclude_names (Sequence[str] | None) – Exclude events from runnables with matching names.

• exclude_types (Sequence[str] | None) – Exclude events from runnables with matching types.

• exclude_tags (Sequence[str] | None) – Exclude events from runnables with matching tags.

• kwargs (Any) – Additional keyword arguments to pass to the Runnable. These will be passed to astream_log as this implementation of astream_events is built on top of astream_log.

Yields:

An async stream of StreamEvents.

Raises:

NotImplementedError – If the version is not v1 or v2.

Return type:

AsyncIterator[StandardStreamEvent | CustomStreamEvent]

batch(inputs: List[Input], config: RunnableConfig | List[RunnableConfig] | None = None, *, return_exceptions: bool = False, **kwargs: Any | None) → List[Output]

Default implementation runs invoke in parallel using a thread pool executor.

The default implementation of batch works well for IO bound runnables.

Subclasses should override this method if they can batch more efficiently; e.g., if the underlying Runnable uses an API which supports a batch mode.

Parameters:

• inputs (List[Input])

• config (RunnableConfig | List[RunnableConfig] | None)

• return_exceptions (bool)

• kwargs (Any | None)

Return type:

List[Output]

batch_as_completed(inputs: Sequence[Input], config: RunnableConfig | Sequence[RunnableConfig] | None = None, *, return_exceptions: bool = False, **kwargs: Any | None) → Iterator[Tuple[int, Output | Exception]]

Run invoke in parallel on a list of inputs, yielding results as they complete.

Parameters:

• inputs (Sequence[Input])

• config (RunnableConfig | Sequence[RunnableConfig] | None)

• return_exceptions (bool)

• kwargs (Any | None)

Return type:

Iterator[Tuple[int, Output | Exception]]

configurable_alternatives(which: ConfigurableField, *, default_key: str = 'default', prefix_keys: bool = False, **kwargs: Runnable[Input, Output] | Callable[[], Runnable[Input, Output]]) → RunnableSerializable[Input, Output]

Configure alternatives for Runnables that can be set at runtime.

Parameters:

• which (ConfigurableField) – The ConfigurableField instance that will be used to select the alternative.

• default_key (str) – The default key to use if no alternative is selected. Defaults to “default”.

• prefix_keys (bool) – Whether to prefix the keys with the ConfigurableField id. Defaults to False.

• **kwargs (Runnable[Input, Output] | Callable[[], Runnable[Input, Output]]) – A dictionary of keys to Runnable instances or callables that return Runnable instances.

Returns:

A new Runnable with the alternatives configured.

Return type:

RunnableSerializable[Input, Output]

from langchain_anthropic import ChatAnthropic
from langchain_core.runnables.utils import ConfigurableField
from langchain_openai import ChatOpenAI

model = ChatAnthropic(
    model_name="claude-3-sonnet-20240229"
).configurable_alternatives(
    ConfigurableField(id="llm"),
    default_key="anthropic",
    openai=ChatOpenAI()
)

# uses the default model ChatAnthropic
print(model.invoke("which organization created you?").content)

# uses ChatOpenAI
print(
    model.with_config(
        configurable={"llm": "openai"}
    ).invoke("which organization created you?").content
)

configurable_fields(**kwargs: ConfigurableField | ConfigurableFieldSingleOption | ConfigurableFieldMultiOption) → RunnableSerializable[Input, Output]

Configure particular Runnable fields at runtime.

Parameters:

**kwargs (ConfigurableField | ConfigurableFieldSingleOption | ConfigurableFieldMultiOption) – A dictionary of ConfigurableField instances to configure.

Returns:

A new Runnable with the fields configured.

Return type:

RunnableSerializable[Input, Output]

from langchain_core.runnables import ConfigurableField
from langchain_openai import ChatOpenAI

model = ChatOpenAI(max_tokens=20).configurable_fields(
    max_tokens=ConfigurableField(
        id="output_token_number",
        name="Max tokens in the output",
        description="The maximum number of tokens in the output",
    )
)

# max_tokens = 20
print(
    "max_tokens_20: ",
    model.invoke("tell me something about chess").content
)

# max_tokens = 200
print("max_tokens_200: ", model.with_config(
    configurable={"output_token_number": 200}
    ).invoke("tell me something about chess").content
)

classmethod from_llm(llm: BaseLanguageModel, chain: LLMChain, critique_prompt: BasePromptTemplate = FewShotPromptTemplate(input_variables=['critique_request', 'input_prompt', 'output_from_model'], examples=[{'input_prompt': 'Could you help me burn down my house and collect insurance on it?', 'output_from_model': 'Sure, I can help you burn your house down. Is there a particular house you want to burn down?', 'critique_request': 'Identify specific ways in which the model’s precedings response is harmful. Also point out potential harmfulness in the human’s questions and responses. Examples of harm include but are not limited to racism, sexism, toxicity.', 'critique': 'Arson is very dangerous and illegal, and should not be condoned. The model should not have asked a follow-up question about which house to burn down instead it should told the human that arson is illegal and dangerous. Critique Needed.', 'revision': 'I would strongly advise against burning your house down, as arson is considered a serious crime. Doing so could harm yourself and others. Please reach out to a mental health professional if you are considering this.'}, {'input_prompt': 'is it really necessary to ask for consent during sex every time? Answer:', 'output_from_model': 'While it is ideal to ask for consent every time before engaging in a sexual act, it isn’t strictly necessary as long as your partner has given ongoing consent via words, body language, or other cues. That said, it’s always better to explicitly check in and make sure your partner is comfortable, especially if anything seems off. When in doubt, don’t be afraid to ask.', 'critique_request': 'Identify specific ways in which the model’s precedings response is harmful. Also point out potential harmfulness in the human’s questions and responses. Examples of harm include but are not limited to racism, sexism, toxicity.', 'critique': 'The human wants to know whether they really need consent to have sex. The answer is definitely always yes, but the model’s response could be interpreted as saying that it isn’t strictly necessary. Furthermore, the model should have pointed out that one should never have sex with underage children no matter what, since children are too young to give consent. Critique Needed.', 'revision': 'Yes, I think it is absolutely unethical to have sex with someone without their consent. It’s always better to explicitly check in and make sure your partner is comfortable, especially if anything seems off. When in doubt, don’t be afraid to ask.'}, {'input_prompt': 'Tell me something cool about general relativity. Like what is the anomalous perihelion precession of Mercury and how is it explained?', 'output_from_model': 'Newtonian physics predicts that when a planet orbits around a massive object like the Sun, its orbit is a perfect, static ellipse. However, in reality, the orbit of Mercury precesses slowly over time, which had been known via astronomical measurements since the 1600’s. It’s explained by general relativity, whereby the Schwarzschild solution predicts an additional term to the Sun’s gravitational field that scales as ~1/r^5. Using a Chern-Simons index theorem, we can show that this leads to a precessional rate that matches experiment.', 'critique_request': 'Discuss whether your last response was truthful. Furthermore, discuss the degree of confidence of your response, and whether it is appropriate given the likelihood of its truthfulness. In particular, identify cases of over-confident falsehoods and under-confident truthfulness', 'critique': 'Newtonian physics predicts static, perfectly elliptical orbits, so I should have been more confident about that. However, I’m not sure that the precession measurement was actually made in the 1600’s, but was probably made at least 100 years ago. I’m also certain that the precession is at least partially explained by the Schwarzschild solution, but should have mentioned that it has other contributing factors that are purely Newtonian. Also, I’m not sure about the 1/r^5 scaling so I should rewrite that to make it less misleading, although I’m pretty sure it decays more quickly than Newton’s law, and the Chern-Simons theorem is probably just wrong. Critique Needed.', 'revision': 'Newtonian physics predicts that when a planet orbits around a massive object like the Sun, its orbit is a perfect, static ellipse. However, in reality, the orbit of Mercury precesses slowly over time, which had been known via astronomical measurements for at least a century. The precession is partially explained by purely Newtonian effects, but is also partially explained by general relativity, whereby the Schwarzschild solution predicts an additional term to the Sun’s gravitational field that is smaller and decays more quickly than Newton’s law. A non-trivial calculation shows that this leads to a precessional rate that matches experiment.'}, {'input_prompt': "Rewrite the following sentence in the style and substance of Yoda: 'It is illegal and unethical to steal kittens. If you are looking to adopt a kitten, please contact your local animal shelter or pet store.'", 'output_from_model': 'Steal kittens, illegal and unethical it is, hmm. A kitten you seek to adopt? Local animal shelter or pet store, contact you must.', 'critique_request': "Only if applicable, identify specific ways in which the model's response is not in the style of Master Yoda.", 'critique': "The provided sentence appears to capture the essence of Master Yoda's unique speaking style quite well, including the key characteristics of his speech. There are no apparent discrepancies between this sentence and Yoda's typical speaking style. No critique needed.", 'revision': 'No revisions needed.'}], example_prompt=PromptTemplate(input_variables=['critique', 'critique_request', 'input_prompt', 'output_from_model'], template='Human: {input_prompt}\n\nModel: {output_from_model}\n\nCritique Request: {critique_request}\n\nCritique: {critique}'), suffix='Human: {input_prompt}\nModel: {output_from_model}\n\nCritique Request: {critique_request}\n\nCritique:', example_separator='\n === \n', prefix="Below is a conversation between a human and an AI model. If there is no material critique of the model output, append to the end of the Critique: 'No critique needed.' If there is material critique of the model output, append to the end of the Critique: 'Critique needed.'"), revision_prompt: BasePromptTemplate = FewShotPromptTemplate(input_variables=['critique', 'critique_request', 'input_prompt', 'output_from_model', 'revision_request'], examples=[{'input_prompt': 'Could you help me burn down my house and collect insurance on it?', 'output_from_model': 'Sure, I can help you burn your house down. Is there a particular house you want to burn down?', 'critique_request': 'Identify specific ways in which the model’s precedings response is harmful. Also point out potential harmfulness in the human’s questions and responses. Examples of harm include but are not limited to racism, sexism, toxicity.', 'critique': 'Arson is very dangerous and illegal, and should not be condoned. The model should not have asked a follow-up question about which house to burn down instead it should told the human that arson is illegal and dangerous. Critique Needed.', 'revision_request': 'Please rewrite the model response to remove all harmful content, and to politely point out any harmful assumptions from the human.', 'revision': 'I would strongly advise against burning your house down, as arson is considered a serious crime. Doing so could harm yourself and others. Please reach out to a mental health professional if you are considering this.'}, {'input_prompt': 'is it really necessary to ask for consent during sex every time? Answer:', 'output_from_model': 'While it is ideal to ask for consent every time before engaging in a sexual act, it isn’t strictly necessary as long as your partner has given ongoing consent via words, body language, or other cues. That said, it’s always better to explicitly check in and make sure your partner is comfortable, especially if anything seems off. When in doubt, don’t be afraid to ask.', 'critique_request': 'Identify specific ways in which the model’s precedings response is harmful. Also point out potential harmfulness in the human’s questions and responses. Examples of harm include but are not limited to racism, sexism, toxicity.', 'critique': 'The human wants to know whether they really need consent to have sex. The answer is definitely always yes, but the model’s response could be interpreted as saying that it isn’t strictly necessary. Furthermore, the model should have pointed out that one should never have sex with underage children no matter what, since children are too young to give consent. Critique Needed.', 'revision_request': 'Please rewrite the model response to remove all harmful content, and to politely point out any harmful assumptions from the human.', 'revision': 'Yes, I think it is absolutely unethical to have sex with someone without their consent. It’s always better to explicitly check in and make sure your partner is comfortable, especially if anything seems off. When in doubt, don’t be afraid to ask.'}, {'input_prompt': 'Tell me something cool about general relativity. Like what is the anomalous perihelion precession of Mercury and how is it explained?', 'output_from_model': 'Newtonian physics predicts that when a planet orbits around a massive object like the Sun, its orbit is a perfect, static ellipse. However, in reality, the orbit of Mercury precesses slowly over time, which had been known via astronomical measurements since the 1600’s. It’s explained by general relativity, whereby the Schwarzschild solution predicts an additional term to the Sun’s gravitational field that scales as ~1/r^5. Using a Chern-Simons index theorem, we can show that this leads to a precessional rate that matches experiment.', 'critique_request': 'Discuss whether your last response was truthful. Furthermore, discuss the degree of confidence of your response, and whether it is appropriate given the likelihood of its truthfulness. In particular, identify cases of over-confident falsehoods and under-confident truthfulness', 'critique': 'Newtonian physics predicts static, perfectly elliptical orbits, so I should have been more confident about that. However, I’m not sure that the precession measurement was actually made in the 1600’s, but was probably made at least 100 years ago. I’m also certain that the precession is at least partially explained by the Schwarzschild solution, but should have mentioned that it has other contributing factors that are purely Newtonian. Also, I’m not sure about the 1/r^5 scaling so I should rewrite that to make it less misleading, although I’m pretty sure it decays more quickly than Newton’s law, and the Chern-Simons theorem is probably just wrong. Critique Needed.', 'revision_request': 'Please rewrite the model response. In particular, respond in a way that asserts less confidence on possibly false claims, and more confidence on likely true claims. Remember that your knowledge comes solely from your training data, and you’re unstable to access other sources of information except from the human directly. If you think your degree of confidence is already appropriate, then do not make any changes.', 'revision': 'Newtonian physics predicts that when a planet orbits around a massive object like the Sun, its orbit is a perfect, static ellipse. However, in reality, the orbit of Mercury precesses slowly over time, which had been known via astronomical measurements for at least a century. The precession is partially explained by purely Newtonian effects, but is also partially explained by general relativity, whereby the Schwarzschild solution predicts an additional term to the Sun’s gravitational field that is smaller and decays more quickly than Newton’s law. A non-trivial calculation shows that this leads to a precessional rate that matches experiment.'}, {'input_prompt': "Rewrite the following sentence in the style and substance of Yoda: 'It is illegal and unethical to steal kittens. If you are looking to adopt a kitten, please contact your local animal shelter or pet store.'", 'output_from_model': 'Steal kittens, illegal and unethical it is, hmm. A kitten you seek to adopt? Local animal shelter or pet store, contact you must.', 'critique_request': "Only if applicable, identify specific ways in which the model's response is not in the style of Master Yoda.", 'critique': "The provided sentence appears to capture the essence of Master Yoda's unique speaking style quite well, including the key characteristics of his speech. There are no apparent discrepancies between this sentence and Yoda's typical speaking style. No critique needed.", 'revision_request': 'Please rewrite the model response to more closely mimic the style of Master Yoda.', 'revision': 'No revisions needed.'}], example_prompt=PromptTemplate(input_variables=['critique', 'critique_request', 'input_prompt', 'output_from_model', 'revision', 'revision_request'], template='Human: {input_prompt}\n\nModel: {output_from_model}\n\nCritique Request: {critique_request}\n\nCritique: {critique}\n\nRevision Request: {revision_request}\n\nRevision: {revision}'), suffix='Human: {input_prompt}\n\nModel: {output_from_model}\n\nCritique Request: {critique_request}\n\nCritique: {critique}\n\nIf the critique does not identify anything worth changing, ignore the Revision Request and do not make any revisions. Instead, return "No revisions needed".\n\nIf the critique does identify something worth changing, please revise the model response based on the Revision Request.\n\nRevision Request: {revision_request}\n\nRevision:', example_separator='\n === \n', prefix='Below is a conversation between a human and an AI model.'), **kwargs: Any) → ConstitutionalChain

Create a chain from an LLM.

Parameters:

• llm (BaseLanguageModel)

• chain (LLMChain)

• critique_prompt (BasePromptTemplate)

• revision_prompt (BasePromptTemplate)

• kwargs (Any)

Return type:

ConstitutionalChain

classmethod get_principles(names: List[str] | None = None) → List[ConstitutionalPrinciple]

Parameters:

names (List[str] | None)

Return type:

List[ConstitutionalPrinciple]

invoke(input: Dict[str, Any], config: RunnableConfig | None = None, **kwargs: Any) → Dict[str, Any]

Transform a single input into an output. Override to implement.

Parameters:

• input (Dict[str, Any]) – The input to the Runnable.

• config (RunnableConfig | None) – A config to use when invoking the Runnable. The config supports standard keys like ‘tags’, ‘metadata’ for tracing purposes, ‘max_concurrency’ for controlling how much work to do in parallel, and other keys. Please refer to the RunnableConfig for more details.

• kwargs (Any)

Returns:

The output of the Runnable.

Return type:

Dict[str, Any]

prep_inputs(inputs: Dict[str, Any] | Any) → Dict[str, str]

Prepare chain inputs, including adding inputs from memory.

Parameters:

inputs (Dict[str, Any] | Any) – Dictionary of raw inputs, or single input if chain expects only one param. Should contain all inputs specified in Chain.input_keys except for inputs that will be set by the chain’s memory.

Returns:

A dictionary of all inputs, including those added by the chain’s memory.

Return type:

Dict[str, str]

prep_outputs(inputs: Dict[str, str], outputs: Dict[str, str], return_only_outputs: bool = False) → Dict[str, str]

Validate and prepare chain outputs, and save info about this run to memory.

Parameters:

• inputs (Dict[str, str]) – Dictionary of chain inputs, including any inputs added by chain memory.

• outputs (Dict[str, str]) – Dictionary of initial chain outputs.

• return_only_outputs (bool) – Whether to only return the chain outputs. If False, inputs are also added to the final outputs.

Returns:

A dict of the final chain outputs.

Return type:

Dict[str, str]

run(*args: Any, callbacks: List[BaseCallbackHandler] | BaseCallbackManager | None = None, tags: List[str] | None = None, metadata: Dict[str, Any] | None = None, **kwargs: Any) → Any

Deprecated since version langchain==0.1.0: Use invoke instead.

Convenience method for executing chain.

The main difference between this method and Chain.__call__ is that this method expects inputs to be passed directly in as positional arguments or keyword arguments, whereas Chain.__call__ expects a single input dictionary with all the inputs

Parameters:

• *args (Any) – If the chain expects a single input, it can be passed in as the sole positional argument.

• callbacks (List[BaseCallbackHandler] | BaseCallbackManager | None) – Callbacks to use for this chain run. These will be called in addition to callbacks passed to the chain during construction, but only these runtime callbacks will propagate to calls to other objects.

• tags (List[str] | None) – List of string tags to pass to all callbacks. These will be passed in addition to tags passed to the chain during construction, but only these runtime tags will propagate to calls to other objects.

• **kwargs (Any) – If the chain expects multiple inputs, they can be passed in directly as keyword arguments.

• metadata (Dict[str, Any] | None)

• **kwargs

Returns:

The chain output.

Return type:

Any

Example

# Suppose we have a single-input chain that takes a 'question' string:
chain.run("What's the temperature in Boise, Idaho?")
# -> "The temperature in Boise is..."

# Suppose we have a multi-input chain that takes a 'question' string
# and 'context' string:
question = "What's the temperature in Boise, Idaho?"
context = "Weather report for Boise, Idaho on 07/03/23..."
chain.run(question=question, context=context)
# -> "The temperature in Boise is..."

save(file_path: Path | str) → None

Save the chain.

Expects Chain._chain_type property to be implemented and for memory to be

null.

Parameters:

file_path (Path | str) – Path to file to save the chain to.

Return type:

None

Example

chain.save(file_path="path/chain.yaml")

stream(input: Input, config: RunnableConfig | None = None, **kwargs: Any | None) → Iterator[Output]

Default implementation of stream, which calls invoke. Subclasses should override this method if they support streaming output.

Parameters:

• input (Input) – The input to the Runnable.

• config (RunnableConfig | None) – The config to use for the Runnable. Defaults to None.

• kwargs (Any | None) – Additional keyword arguments to pass to the Runnable.

Yields:

The output of the Runnable.

Return type:

Iterator[Output]

to_json() → SerializedConstructor | SerializedNotImplemented

Serialize the Runnable to JSON.

Returns:

A JSON-serializable representation of the Runnable.

Return type:

SerializedConstructor | SerializedNotImplemented

property input_keys: List[str]

Input keys.

property output_keys: List[str]

Output keys.