github/langchain
1
0
Fork 0
mirror of https://github.com/hwchase17/langchain.git synced 2025-07-01 10:54:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add openai v2 adapter (#14063)

Browse Source 
### Description

Starting from [openai version
1.0.0](17ac677995 (module-level-client)),
the camel case form of `openai.ChatCompletion` is no longer supported
and has been changed to lowercase `openai.chat.completions`. In
addition, the returned object only accepts attribute access instead of
index access:

```python
import openai

# optional; defaults to `os.environ['OPENAI_API_KEY']`
openai.api_key = '...'

# all client options can be configured just like the `OpenAI` instantiation counterpart
openai.base_url = "https://..."
openai.default_headers = {"x-foo": "true"}

completion = openai.chat.completions.create(
    model="gpt-4",
    messages=[
        {
            "role": "user",
            "content": "How do I output all files in a directory using Python?",
        },
    ],
)
print(completion.choices[0].message.content)
```

So I implemented a compatible adapter that supports both attribute
access and index access:

```python
In [1]: from langchain.adapters import openai as lc_openai
   ...: messages = [{"role": "user", "content": "hi"}]

In [2]: result = lc_openai.chat.completions.create(
   ...:     messages=messages, model="gpt-3.5-turbo", temperature=0
   ...: )

In [3]: result.choices[0].message
Out[3]: {'role': 'assistant', 'content': 'Hello! How can I assist you today?'}

In [4]: result["choices"][0]["message"]
Out[4]: {'role': 'assistant', 'content': 'Hello! How can I assist you today?'}

In [5]: result = await lc_openai.chat.completions.acreate(
   ...:     messages=messages, model="gpt-3.5-turbo", temperature=0
   ...: )

In [6]: result.choices[0].message
Out[6]: {'role': 'assistant', 'content': 'Hello! How can I assist you today?'}

In [7]: result["choices"][0]["message"]
Out[7]: {'role': 'assistant', 'content': 'Hello! How can I assist you today?'}

In [8]: for rs in lc_openai.chat.completions.create(
    ...:     messages=messages, model="gpt-3.5-turbo", temperature=0, stream=True
    ...: ):
    ...:     print(rs.choices[0].delta)
    ...:     print(rs["choices"][0]["delta"])
    ...:
{'role': 'assistant', 'content': ''}
{'role': 'assistant', 'content': ''}
{'content': 'Hello'}
{'content': 'Hello'}
{'content': '!'}
{'content': '!'}

In [20]: async for rs in await lc_openai.chat.completions.acreate(
    ...:     messages=messages, model="gpt-3.5-turbo", temperature=0, stream=True
    ...: ):
    ...:     print(rs.choices[0].delta)
    ...:     print(rs["choices"][0]["delta"])
    ...:
{'role': 'assistant', 'content': ''}
{'role': 'assistant', 'content': ''}
{'content': 'Hello'}
{'content': 'Hello'}
{'content': '!'}
{'content': '!'}
...
```

### Twitter handle

[lin_bob57617](https://twitter.com/lin_bob57617)
This commit is contained in:
Bob Lin 2023-12-04 14:12:30 -06:00 committed by GitHub
parent 0f02081392
commit cd2028288e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 495 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

285
docs/docs/integrations/adapters/openai-old.ipynb Normal file
View File

@ -0,0 +1,285 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "700a516b",
"metadata": {},
"source": [
"# OpenAI Adapter(Old)\n",
"\n",
"**Please ensure OpenAI library is less than 1.0.0; otherwise, refer to the newer doc [OpenAI Adapter](./openai.ipynb).**\n",
"\n",
"A lot of people get started with OpenAI but want to explore other models. LangChain's integrations with many model providers make this easy to do so. While LangChain has it's own message and model APIs, we've also made it as easy as possible to explore other models by exposing an adapter to adapt LangChain models to the OpenAI api.\n",
"\n",
"At the moment this only deals with output and does not return other information (token counts, stop reasons, etc)."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "6017f26a",
"metadata": {},
"outputs": [],
"source": [
"import openai\n",
"from langchain.adapters import openai as lc_openai"
]
},
{
"cell_type": "markdown",
"id": "b522ceda",
"metadata": {},
"source": [
"## ChatCompletion.create"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "1d22eb61",
"metadata": {},
"outputs": [],
"source": [
"messages = [{\"role\": \"user\", \"content\": \"hi\"}]"
]
},
{
"cell_type": "markdown",
"id": "d550d3ad",
"metadata": {},
"source": [
"Original OpenAI call"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "012d81ae",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': 'Hello! How can I assist you today?'}"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"result = openai.ChatCompletion.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0\n",
")\n",
"result[\"choices\"][0][\"message\"].to_dict_recursive()"
]
},
{
"cell_type": "markdown",
"id": "db5b5500",
"metadata": {},
"source": [
"LangChain OpenAI wrapper call"
]
},
{
"cell_type": "code",
"execution_count": 17,
"id": "c67a5ac8",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': 'Hello! How can I assist you today?'}"
]
},
"execution_count": 17,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"lc_result = lc_openai.ChatCompletion.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0\n",
")\n",
"lc_result[\"choices\"][0][\"message\"]"
]
},
{
"cell_type": "markdown",
"id": "034ba845",
"metadata": {},
"source": [
"Swapping out model providers"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "f7c94827",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': ' Hello!'}"
]
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"lc_result = lc_openai.ChatCompletion.create(\n",
" messages=messages, model=\"claude-2\", temperature=0, provider=\"ChatAnthropic\"\n",
")\n",
"lc_result[\"choices\"][0][\"message\"]"
]
},
{
"cell_type": "markdown",
"id": "cb3f181d",
"metadata": {},
"source": [
"## ChatCompletion.stream"
]
},
{
"cell_type": "markdown",
"id": "f7b8cd18",
"metadata": {},
"source": [
"Original OpenAI call"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "fd8cb1ea",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'role': 'assistant', 'content': ''}\n",
"{'content': 'Hello'}\n",
"{'content': '!'}\n",
"{'content': ' How'}\n",
"{'content': ' can'}\n",
"{'content': ' I'}\n",
"{'content': ' assist'}\n",
"{'content': ' you'}\n",
"{'content': ' today'}\n",
"{'content': '?'}\n",
"{}\n"
]
}
],
"source": [
"for c in openai.ChatCompletion.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0, stream=True\n",
"):\n",
" print(c[\"choices\"][0][\"delta\"].to_dict_recursive())"
]
},
{
"cell_type": "markdown",
"id": "0b2a076b",
"metadata": {},
"source": [
"LangChain OpenAI wrapper call"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "9521218c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'role': 'assistant', 'content': ''}\n",
"{'content': 'Hello'}\n",
"{'content': '!'}\n",
"{'content': ' How'}\n",
"{'content': ' can'}\n",
"{'content': ' I'}\n",
"{'content': ' assist'}\n",
"{'content': ' you'}\n",
"{'content': ' today'}\n",
"{'content': '?'}\n",
"{}\n"
]
}
],
"source": [
"for c in lc_openai.ChatCompletion.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0, stream=True\n",
"):\n",
" print(c[\"choices\"][0][\"delta\"])"
]
},
{
"cell_type": "markdown",
"id": "0fc39750",
"metadata": {},
"source": [
"Swapping out model providers"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "68f0214e",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'role': 'assistant', 'content': ' Hello'}\n",
"{'content': '!'}\n",
"{}\n"
]
}
],
"source": [
"for c in lc_openai.ChatCompletion.create(\n",
" messages=messages,\n",
" model=\"claude-2\",\n",
" temperature=0,\n",
" stream=True,\n",
" provider=\"ChatAnthropic\",\n",
"):\n",
" print(c[\"choices\"][0][\"delta\"])"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.5"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

113
docs/docs/integrations/adapters/openai.ipynb
View File

@ -7,6 +7,8 @@
"source": [
"# OpenAI Adapter\n",
"\n",
"**Please ensure OpenAI library is version 1.0.0 or higher; otherwise, refer to the older doc [OpenAI Adapter(Old)](./openai-old.ipynb).**\n",
"\n",
"A lot of people get started with OpenAI but want to explore other models. LangChain's integrations with many model providers make this easy to do so. While LangChain has it's own message and model APIs, we've also made it as easy as possible to explore other models by exposing an adapter to adapt LangChain models to the OpenAI api.\n",
"\n",
"At the moment this only deals with output and does not return other information (token counts, stop reasons, etc)."
@ -28,12 +30,12 @@
"id": "b522ceda",
"metadata": {},
"source": [
"## ChatCompletion.create"
"## chat.completions.create"
]
},
{
"cell_type": "code",
"execution_count": 29,
"execution_count": 2,
"id": "1d22eb61",
"metadata": {},
"outputs": [],
@ -51,26 +53,29 @@
},
{
"cell_type": "code",
"execution_count": 15,
"execution_count": 3,
"id": "012d81ae",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': 'Hello! How can I assist you today?'}"
"{'content': 'Hello! How can I assist you today?',\n",
" 'role': 'assistant',\n",
" 'function_call': None,\n",
" 'tool_calls': None}"
]
},
"execution_count": 15,
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"result = openai.ChatCompletion.create(\n",
"result = openai.chat.completions.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0\n",
")\n",
"result[\"choices\"][0][\"message\"].to_dict_recursive()"
"result.choices[0].message.model_dump()"
]
},
{
@ -83,26 +88,48 @@
},
{
"cell_type": "code",
"execution_count": 17,
"execution_count": 4,
"id": "c67a5ac8",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': 'Hello! How can I assist you today?'}"
"{'role': 'assistant', 'content': 'Hello! How can I help you today?'}"
]
},
"execution_count": 17,
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"lc_result = lc_openai.ChatCompletion.create(\n",
"lc_result = lc_openai.chat.completions.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0\n",
")\n",
"lc_result[\"choices\"][0][\"message\"]"
"\n",
"lc_result.choices[0].message # Attribute access"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "37a6e461-8608-47f6-ac45-12ad753c062a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': 'Hello! How can I help you today?'}"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"lc_result[\"choices\"][0][\"message\"] # Also compatible with index access"
]
},
{
@ -115,26 +142,26 @@
},
{
"cell_type": "code",
"execution_count": 19,
"execution_count": 6,
"id": "f7c94827",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'role': 'assistant', 'content': ' Hello!'}"
"{'role': 'assistant', 'content': 'Hello! How can I assist you today?'}"
]
},
"execution_count": 19,
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"lc_result = lc_openai.ChatCompletion.create(\n",
"lc_result = lc_openai.chat.completions.create(\n",
" messages=messages, model=\"claude-2\", temperature=0, provider=\"ChatAnthropic\"\n",
")\n",
"lc_result[\"choices\"][0][\"message\"]"
"lc_result.choices[0].message"
]
},
{
@ -142,7 +169,7 @@
"id": "cb3f181d",
"metadata": {},
"source": [
"## ChatCompletion.stream"
"## chat.completions.stream"
]
},
{
@ -155,7 +182,7 @@
},
{
"cell_type": "code",
"execution_count": 24,
"execution_count": 7,
"id": "fd8cb1ea",
"metadata": {},
"outputs": [
@ -163,25 +190,25 @@
"name": "stdout",
"output_type": "stream",
"text": [
"{'role': 'assistant', 'content': ''}\n",
"{'content': 'Hello'}\n",
"{'content': '!'}\n",
"{'content': ' How'}\n",
"{'content': ' can'}\n",
"{'content': ' I'}\n",
"{'content': ' assist'}\n",
"{'content': ' you'}\n",
"{'content': ' today'}\n",
"{'content': '?'}\n",
"{}\n"
"{'content': '', 'function_call': None, 'role': 'assistant', 'tool_calls': None}\n",
"{'content': 'Hello', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': '!', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': ' How', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': ' can', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': ' I', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': ' assist', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': ' you', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': ' today', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': '?', 'function_call': None, 'role': None, 'tool_calls': None}\n",
"{'content': None, 'function_call': None, 'role': None, 'tool_calls': None}\n"
]
}
],
"source": [
"for c in openai.ChatCompletion.create(\n",
"for c in openai.chat.completions.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0, stream=True\n",
"):\n",
" print(c[\"choices\"][0][\"delta\"].to_dict_recursive())"
" print(c.choices[0].delta.model_dump())"
]
},
{
@ -194,7 +221,7 @@
},
{
"cell_type": "code",
"execution_count": 30,
"execution_count": 8,
"id": "9521218c",
"metadata": {},
"outputs": [
@ -217,10 +244,10 @@
}
],
"source": [
"for c in lc_openai.ChatCompletion.create(\n",
"for c in lc_openai.chat.completions.create(\n",
" messages=messages, model=\"gpt-3.5-turbo\", temperature=0, stream=True\n",
"):\n",
" print(c[\"choices\"][0][\"delta\"])"
" print(c.choices[0].delta)"
]
},
{
@ -233,7 +260,7 @@
},
{
"cell_type": "code",
"execution_count": 31,
"execution_count": 9,
"id": "68f0214e",
"metadata": {},
"outputs": [
@ -241,14 +268,22 @@
"name": "stdout",
"output_type": "stream",
"text": [
"{'role': 'assistant', 'content': ' Hello'}\n",
"{'role': 'assistant', 'content': ''}\n",
"{'content': 'Hello'}\n",
"{'content': '!'}\n",
"{'content': ' How'}\n",
"{'content': ' can'}\n",
"{'content': ' I'}\n",
"{'content': ' assist'}\n",
"{'content': ' you'}\n",
"{'content': ' today'}\n",
"{'content': '?'}\n",
"{}\n"
]
}
],
"source": [
"for c in lc_openai.ChatCompletion.create(\n",
"for c in lc_openai.chat.completions.create(\n",
" messages=messages,\n",
" model=\"claude-2\",\n",
" temperature=0,\n",
@ -275,7 +310,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.12"
"version": "3.11.5"
}
},
"nbformat": 4,

137
libs/langchain/langchain/adapters/openai.py
View File

@ -25,6 +25,7 @@ from langchain_core.messages import (
SystemMessage,
ToolMessage,
)
from langchain_core.pydantic_v1 import BaseModel
from typing_extensions import Literal
@ -38,6 +39,29 @@ async def aenumerate(
i += 1
class IndexableBaseModel(BaseModel):
"""Allows a BaseModel to return its fields by string variable indexing"""
def __getitem__(self, item: str) -> Any:
return getattr(self, item)
class Choice(IndexableBaseModel):
message: dict
class ChatCompletions(IndexableBaseModel):
choices: List[Choice]
class ChoiceChunk(IndexableBaseModel):
delta: dict
class ChatCompletionChunk(IndexableBaseModel):
choices: List[ChoiceChunk]
def convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
"""Convert a dictionary to a LangChain message.
@ -129,7 +153,7 @@ def convert_openai_messages(messages: Sequence[Dict[str, Any]]) -> List[BaseMess
return [convert_dict_to_message(m) for m in messages]
def _convert_message_chunk_to_delta(chunk: BaseMessageChunk, i: int) -> Dict[str, Any]:
def _convert_message_chunk(chunk: BaseMessageChunk, i: int) -> dict:
_dict: Dict[str, Any] = {}
if isinstance(chunk, AIMessageChunk):
if i == 0:
@ -148,6 +172,11 @@ def _convert_message_chunk_to_delta(chunk: BaseMessageChunk, i: int) -> Dict[str
# This only happens at the end of streams, and OpenAI returns as empty dict
if _dict == {"content": ""}:
_dict = {}
return _dict
def _convert_message_chunk_to_delta(chunk: BaseMessageChunk, i: int) -> Dict[str, Any]:
_dict = _convert_message_chunk(chunk, i)
return {"choices": [{"delta": _dict}]}
@ -262,3 +291,109 @@ def convert_messages_for_finetuning(
for session in sessions
if _has_assistant_message(session)
]
class Completions:
"""Completion."""
@overload
@staticmethod
def create(
messages: Sequence[Dict[str, Any]],
*,
provider: str = "ChatOpenAI",
stream: Literal[False] = False,
**kwargs: Any,
) -> ChatCompletions:
...
@overload
@staticmethod
def create(
messages: Sequence[Dict[str, Any]],
*,
provider: str = "ChatOpenAI",
stream: Literal[True],
**kwargs: Any,
) -> Iterable:
...
@staticmethod
def create(
messages: Sequence[Dict[str, Any]],
*,
provider: str = "ChatOpenAI",
stream: bool = False,
**kwargs: Any,
) -> Union[ChatCompletions, Iterable]:
models = importlib.import_module("langchain.chat_models")
model_cls = getattr(models, provider)
model_config = model_cls(**kwargs)
converted_messages = convert_openai_messages(messages)
if not stream:
result = model_config.invoke(converted_messages)
return ChatCompletions(
choices=[Choice(message=convert_message_to_dict(result))]
)
else:
return (
ChatCompletionChunk(
choices=[ChoiceChunk(delta=_convert_message_chunk(c, i))]
)
for i, c in enumerate(model_config.stream(converted_messages))
)
@overload
@staticmethod
async def acreate(
messages: Sequence[Dict[str, Any]],
*,
provider: str = "ChatOpenAI",
stream: Literal[False] = False,
**kwargs: Any,
) -> ChatCompletions:
...
@overload
@staticmethod
async def acreate(
messages: Sequence[Dict[str, Any]],
*,
provider: str = "ChatOpenAI",
stream: Literal[True],
**kwargs: Any,
) -> AsyncIterator:
...
@staticmethod
async def acreate(
messages: Sequence[Dict[str, Any]],
*,
provider: str = "ChatOpenAI",
stream: bool = False,
**kwargs: Any,
) -> Union[ChatCompletions, AsyncIterator]:
models = importlib.import_module("langchain.chat_models")
model_cls = getattr(models, provider)
model_config = model_cls(**kwargs)
converted_messages = convert_openai_messages(messages)
if not stream:
result = await model_config.ainvoke(converted_messages)
return ChatCompletions(
choices=[Choice(message=convert_message_to_dict(result))]
)
else:
return (
ChatCompletionChunk(
choices=[ChoiceChunk(delta=_convert_message_chunk(c, i))]
)
async for i, c in aenumerate(model_config.astream(converted_messages))
)
class Chat:
def __init__(self) -> None:
self.completions = Completions()
chat = Chat()