ChatOutlines

This will help you get started with Outlines chat models. For detailed documentation of all ChatOutlines features and configurations head to the API reference.

Outlines is a library for constrained language generation. It allows you to use large language models (LLMs) with various backends while applying constraints to the generated output.

Overview

Integration details

Class	Package	Local	Serializable	JS support	Downloads	Version
ChatOutlines	langchain-community	✅	❌	❌

Model features

Tool calling	Structured output	JSON mode	Image input	Audio input	Video input	Token-level streaming	Native async	Token usage	Logprobs
✅	✅	✅	✅	❌	❌	✅	❌	❌	❌

Setup

To access Outlines models you’ll need to have an internet connection to download the model weights from huggingface. Depending on the backend you need to install the required dependencies (see Outlines docs)

Credentials

There is no built-in auth mechanism for Outlines.

Installation

The LangChain Outlines integration lives in the langchain-community package and requires the outlines library:

%pip install -qU langchain-community outlines

Instantiation

Now we can instantiate our model object and generate chat completions:

from langchain_community.chat_models.outlines import ChatOutlines

# For llamacpp backend
model = ChatOutlines(model="TheBloke/phi-2-GGUF/phi-2.Q4_K_M.gguf", backend="llamacpp")

# For vllm backend (not available on Mac)
model = ChatOutlines(model="meta-llama/Llama-3.2-1B", backend="vllm")

# For mlxlm backend (only available on Mac)
model = ChatOutlines(model="mistralai/Ministral-8B-Instruct-2410", backend="mlxlm")

# For huggingface transformers backend
model = ChatOutlines(model="microsoft/phi-2")  # defaults to transformers backend

Invocation

from langchain_core.messages import HumanMessage

messages = [HumanMessage(content="What will the capital of mars be called?")]
response = model.invoke(messages)

response.content

Streaming

ChatOutlines supports streaming of tokens:

messages = [HumanMessage(content="Count to 10 in French:")]

for chunk in model.stream(messages):
    print(chunk.content, end="", flush=True)

Chaining

from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a helpful assistant that translates {input_language} to {output_language}.",
        ),
        ("human", "{input}"),
    ]
)

chain = prompt | model
chain.invoke(
    {
        "input_language": "English",
        "output_language": "German",
        "input": "I love programming.",
    }
)

Constrained Generation

ChatOutlines allows you to apply various constraints to the generated output:

Regex Constraint

model.regex = r"((25[0-5]|2[0-4]\d|[01]?\d\d?)\.){3}(25[0-5]|2[0-4]\d|[01]?\d\d?)"

response = model.invoke("What is the IP address of Google's DNS server?")

response.content

Type Constraints

model.type_constraints = int
response = model.invoke("What is the answer to life, the universe, and everything?")

response.content

Pydantic and JSON Schemas

from pydantic import BaseModel


class Person(BaseModel):
    name: str


model.json_schema = Person
response = model.invoke("Who are the main contributors to LangChain?")
person = Person.model_validate_json(response.content)

person

Context Free Grammars

model.grammar = """
?start: expression
?expression: term (("+" | "-") term)*
?term: factor (("*" | "/") factor)*
?factor: NUMBER | "-" factor | "(" expression ")"
%import common.NUMBER
%import common.WS
%ignore WS
"""
response = model.invoke("Give me a complex arithmetic expression:")

response.content

LangChain’s Structured Output

You can also use LangChain’s Structured Output with ChatOutlines:

from pydantic import BaseModel


class AnswerWithJustification(BaseModel):
    answer: str
    justification: str


_model = model.with_structured_output(AnswerWithJustification)
result = _model.invoke("What weighs more, a pound of bricks or a pound of feathers?")

result

API reference

For detailed documentation of all ChatOutlines features and configurations head to the API reference: python.langchain.com/api_reference/community/chat_models/langchain_community.chat_models.outlines.ChatOutlines.html

Full Outlines Documentation

dottxt-ai.github.io/outlines/latest/