langchain/libs/community/tests/integration_tests/chat_models/test_openai.py
Erick Friis 5f839beab9
community: replace deprecated davinci models (#14860)
This is technically a breaking change because it'll switch out default
models from `text-davinci-003` to `gpt-3.5-turbo-instruct`, but OpenAI
is shutting off those endpoints on 1/4 anyways.

Feels less disruptive to switch out the default instead.
2023-12-18 13:49:46 -08:00

333 lines
11 KiB
Python

"""Test ChatOpenAI wrapper."""
from typing import Any, Optional
import pytest
from langchain_core.callbacks import CallbackManager
from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
from langchain_core.outputs import (
ChatGeneration,
ChatResult,
LLMResult,
)
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.pydantic_v1 import BaseModel, Field
from langchain_community.chat_models.openai import ChatOpenAI
from tests.unit_tests.callbacks.fake_callback_handler import FakeCallbackHandler
@pytest.mark.scheduled
def test_chat_openai() -> None:
"""Test ChatOpenAI wrapper."""
chat = ChatOpenAI(
temperature=0.7,
base_url=None,
organization=None,
openai_proxy=None,
timeout=10.0,
max_retries=3,
http_client=None,
n=1,
max_tokens=10,
default_headers=None,
default_query=None,
)
message = HumanMessage(content="Hello")
response = chat([message])
assert isinstance(response, BaseMessage)
assert isinstance(response.content, str)
def test_chat_openai_model() -> None:
"""Test ChatOpenAI wrapper handles model_name."""
chat = ChatOpenAI(model="foo")
assert chat.model_name == "foo"
chat = ChatOpenAI(model_name="bar")
assert chat.model_name == "bar"
def test_chat_openai_system_message() -> None:
"""Test ChatOpenAI wrapper with system message."""
chat = ChatOpenAI(max_tokens=10)
system_message = SystemMessage(content="You are to chat with the user.")
human_message = HumanMessage(content="Hello")
response = chat([system_message, human_message])
assert isinstance(response, BaseMessage)
assert isinstance(response.content, str)
@pytest.mark.scheduled
def test_chat_openai_generate() -> None:
"""Test ChatOpenAI wrapper with generate."""
chat = ChatOpenAI(max_tokens=10, n=2)
message = HumanMessage(content="Hello")
response = chat.generate([[message], [message]])
assert isinstance(response, LLMResult)
assert len(response.generations) == 2
assert response.llm_output
for generations in response.generations:
assert len(generations) == 2
for generation in generations:
assert isinstance(generation, ChatGeneration)
assert isinstance(generation.text, str)
assert generation.text == generation.message.content
@pytest.mark.scheduled
def test_chat_openai_multiple_completions() -> None:
"""Test ChatOpenAI wrapper with multiple completions."""
chat = ChatOpenAI(max_tokens=10, n=5)
message = HumanMessage(content="Hello")
response = chat._generate([message])
assert isinstance(response, ChatResult)
assert len(response.generations) == 5
for generation in response.generations:
assert isinstance(generation.message, BaseMessage)
assert isinstance(generation.message.content, str)
@pytest.mark.scheduled
def test_chat_openai_streaming() -> None:
"""Test that streaming correctly invokes on_llm_new_token callback."""
callback_handler = FakeCallbackHandler()
callback_manager = CallbackManager([callback_handler])
chat = ChatOpenAI(
max_tokens=10,
streaming=True,
temperature=0,
callback_manager=callback_manager,
verbose=True,
)
message = HumanMessage(content="Hello")
response = chat([message])
assert callback_handler.llm_streams > 0
assert isinstance(response, BaseMessage)
@pytest.mark.scheduled
def test_chat_openai_streaming_generation_info() -> None:
"""Test that generation info is preserved when streaming."""
class _FakeCallback(FakeCallbackHandler):
saved_things: dict = {}
def on_llm_end(
self,
*args: Any,
**kwargs: Any,
) -> Any:
# Save the generation
self.saved_things["generation"] = args[0]
callback = _FakeCallback()
callback_manager = CallbackManager([callback])
chat = ChatOpenAI(
max_tokens=2,
temperature=0,
callback_manager=callback_manager,
)
list(chat.stream("hi"))
generation = callback.saved_things["generation"]
# `Hello!` is two tokens, assert that that is what is returned
assert generation.generations[0][0].text == "Hello!"
def test_chat_openai_llm_output_contains_model_name() -> None:
"""Test llm_output contains model_name."""
chat = ChatOpenAI(max_tokens=10)
message = HumanMessage(content="Hello")
llm_result = chat.generate([[message]])
assert llm_result.llm_output is not None
assert llm_result.llm_output["model_name"] == chat.model_name
def test_chat_openai_streaming_llm_output_contains_model_name() -> None:
"""Test llm_output contains model_name."""
chat = ChatOpenAI(max_tokens=10, streaming=True)
message = HumanMessage(content="Hello")
llm_result = chat.generate([[message]])
assert llm_result.llm_output is not None
assert llm_result.llm_output["model_name"] == chat.model_name
def test_chat_openai_invalid_streaming_params() -> None:
"""Test that streaming correctly invokes on_llm_new_token callback."""
with pytest.raises(ValueError):
ChatOpenAI(
max_tokens=10,
streaming=True,
temperature=0,
n=5,
)
@pytest.mark.scheduled
async def test_async_chat_openai() -> None:
"""Test async generation."""
chat = ChatOpenAI(max_tokens=10, n=2)
message = HumanMessage(content="Hello")
response = await chat.agenerate([[message], [message]])
assert isinstance(response, LLMResult)
assert len(response.generations) == 2
assert response.llm_output
for generations in response.generations:
assert len(generations) == 2
for generation in generations:
assert isinstance(generation, ChatGeneration)
assert isinstance(generation.text, str)
assert generation.text == generation.message.content
@pytest.mark.scheduled
async def test_async_chat_openai_streaming() -> None:
"""Test that streaming correctly invokes on_llm_new_token callback."""
callback_handler = FakeCallbackHandler()
callback_manager = CallbackManager([callback_handler])
chat = ChatOpenAI(
max_tokens=10,
streaming=True,
temperature=0,
callback_manager=callback_manager,
verbose=True,
)
message = HumanMessage(content="Hello")
response = await chat.agenerate([[message], [message]])
assert callback_handler.llm_streams > 0
assert isinstance(response, LLMResult)
assert len(response.generations) == 2
for generations in response.generations:
assert len(generations) == 1
for generation in generations:
assert isinstance(generation, ChatGeneration)
assert isinstance(generation.text, str)
assert generation.text == generation.message.content
@pytest.mark.scheduled
async def test_async_chat_openai_bind_functions() -> None:
"""Test ChatOpenAI wrapper with multiple completions."""
class Person(BaseModel):
"""Identifying information about a person."""
name: str = Field(..., title="Name", description="The person's name")
age: int = Field(..., title="Age", description="The person's age")
fav_food: Optional[str] = Field(
default=None, title="Fav Food", description="The person's favorite food"
)
chat = ChatOpenAI(
max_tokens=30,
n=1,
streaming=True,
).bind_functions(functions=[Person], function_call="Person")
prompt = ChatPromptTemplate.from_messages(
[
("system", "Use the provided Person function"),
("user", "{input}"),
]
)
chain = prompt | chat
message = HumanMessage(content="Sally is 13 years old")
response = await chain.abatch([{"input": message}])
assert isinstance(response, list)
assert len(response) == 1
for generation in response:
assert isinstance(generation, AIMessage)
def test_chat_openai_extra_kwargs() -> None:
"""Test extra kwargs to chat openai."""
# Check that foo is saved in extra_kwargs.
llm = ChatOpenAI(foo=3, max_tokens=10)
assert llm.max_tokens == 10
assert llm.model_kwargs == {"foo": 3}
# Test that if extra_kwargs are provided, they are added to it.
llm = ChatOpenAI(foo=3, model_kwargs={"bar": 2})
assert llm.model_kwargs == {"foo": 3, "bar": 2}
# Test that if provided twice it errors
with pytest.raises(ValueError):
ChatOpenAI(foo=3, model_kwargs={"foo": 2})
# Test that if explicit param is specified in kwargs it errors
with pytest.raises(ValueError):
ChatOpenAI(model_kwargs={"temperature": 0.2})
# Test that "model" cannot be specified in kwargs
with pytest.raises(ValueError):
ChatOpenAI(model_kwargs={"model": "gpt-3.5-turbo-instruct"})
@pytest.mark.scheduled
def test_openai_streaming() -> None:
"""Test streaming tokens from OpenAI."""
llm = ChatOpenAI(max_tokens=10)
for token in llm.stream("I'm Pickle Rick"):
assert isinstance(token.content, str)
@pytest.mark.scheduled
async def test_openai_astream() -> None:
"""Test streaming tokens from OpenAI."""
llm = ChatOpenAI(max_tokens=10)
async for token in llm.astream("I'm Pickle Rick"):
assert isinstance(token.content, str)
@pytest.mark.scheduled
async def test_openai_abatch() -> None:
"""Test streaming tokens from ChatOpenAI."""
llm = ChatOpenAI(max_tokens=10)
result = await llm.abatch(["I'm Pickle Rick", "I'm not Pickle Rick"])
for token in result:
assert isinstance(token.content, str)
@pytest.mark.scheduled
async def test_openai_abatch_tags() -> None:
"""Test batch tokens from ChatOpenAI."""
llm = ChatOpenAI(max_tokens=10)
result = await llm.abatch(
["I'm Pickle Rick", "I'm not Pickle Rick"], config={"tags": ["foo"]}
)
for token in result:
assert isinstance(token.content, str)
@pytest.mark.scheduled
def test_openai_batch() -> None:
"""Test batch tokens from ChatOpenAI."""
llm = ChatOpenAI(max_tokens=10)
result = llm.batch(["I'm Pickle Rick", "I'm not Pickle Rick"])
for token in result:
assert isinstance(token.content, str)
@pytest.mark.scheduled
async def test_openai_ainvoke() -> None:
"""Test invoke tokens from ChatOpenAI."""
llm = ChatOpenAI(max_tokens=10)
result = await llm.ainvoke("I'm Pickle Rick", config={"tags": ["foo"]})
assert isinstance(result.content, str)
@pytest.mark.scheduled
def test_openai_invoke() -> None:
"""Test invoke tokens from ChatOpenAI."""
llm = ChatOpenAI(max_tokens=10)
result = llm.invoke("I'm Pickle Rick", config=dict(tags=["foo"]))
assert isinstance(result.content, str)