Fix langchain.llms OpenAI completion doesn't work due to v1 client update

[dosuken123]

This commit fixes the issue that langchain.llms OpenAI completion stopped working since the V1 openai client update.

Replace this entire comment with:

• Description: This PR fixes the issue AttributeError: module 'openai' has no attribute 'Completion' similar to 8e0cb2e and oai v1 embeddings #12969,
• Issue: AttributeError: module 'openai' has no attribute 'Completion' #12967,
• Dependencies: openai v1.x.x client,
• Tag maintainer: @baskaryan,
• Twitter handle: @dosuken123

Please make sure your PR is passing linting and testing before submitting. Run make format, make lint and make test to check this locally.

See contribution guidelines for more information on how to write/run tests, lint, etc:
https://github.com/langchain-ai/langchain/blob/master/.github/CONTRIBUTING.md

If you're adding a new integration, please include:

1. a test for the integration, preferably unit tests that do not rely on network access,
2. an example notebook showing its use. It lives in docs/extras directory.

If no one reviews your PR within a few days, please @-mention one of @baskaryan, @eyurtsev, @hwchase17.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Ignored Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Nov 9, 2023 11:07pm

[dosuken123]

This is the second attempt after I closed #12973.

This is similar fix with 8e0cb2e and #12969.

[dosuken123]

Confirmed locally that it's working with the following script:

from langchain.llms import OpenAI

print(OpenAI().predict("Who's the president of the united states?"))
#=> The President of the United States is Joe Biden.

[dosuken123]

@baskaryan, @eyurtsev, @hwchase17 Would someone have time to review this PR that fixes langchain.llms OpenAI bug?

[lesong36]

@dosuken123 May I know which version of langchain and openai you are using?
I still have this issue for llm and chatmodel is ok.

---

Name: langchain
Version: 0.0.332
Summary: Building applications with LLMs through composability
Home-page: https://github.com/langchain-ai/langchain
Author:
Author-email:
License: MIT
Location: /home/coty/miniconda3/envs/jupyter/lib/python3.11/site-packages
Requires: aiohttp, anyio, dataclasses-json, jsonpatch, langsmith, numpy, pydantic, PyYAML, requests, SQLAlchemy, tenacity
Required-by: jupyter_ai, jupyter_ai_magics, langchain-experimental, langserve

Name: openai
Version: 1.2.0
Summary: Client library for the openai API
Home-page:
Author:
Author-email: OpenAI support@openai.com
License:
Location: /home/coty/miniconda3/envs/jupyter/lib/python3.11/site-packages
Requires: anyio, distro, httpx, pydantic, tqdm, typing-extensions
Required-by: guidance, jupyter_ai, pyautogen

from langchain.chat_models import ChatOpenAI from langchain.schema.messages import HumanMessage, SystemMessage messages = [ SystemMessage(content="You're a helpful assistant"), HumanMessage(content="Who's the president of the united states?"), ] print(ChatOpenAI().invoke(messages).content)

#As of my knowledge update in May 2021, the President of the United States is Joe Biden. He was inaugurated on January 20, 2021.

from langchain.llms import OpenAI print(OpenAI().invoke("Who's the president of the united states?"))

---

AttributeError Traceback (most recent call last)
Cell In[15], line 2
1 from langchain.llms import OpenAI
----> 2 print(OpenAI().invoke("Who's the president of the united states?"))

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/base.py:229, in BaseLLM.invoke(self, input, config, stop, **kwargs)
219 def invoke(
220 self,
221 input: LanguageModelInput,
(...)
225 **kwargs: Any,
226 ) -> str:
227 config = config or {}
228 return (
--> 229 self.generate_prompt(
230 [self._convert_input(input)],
231 stop=stop,
232 callbacks=config.get("callbacks"),
233 tags=config.get("tags"),
234 metadata=config.get("metadata"),
235 run_name=config.get("run_name"),
236 **kwargs,
237 )
238 .generations[0][0]
239 .text
240 )

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/base.py:507, in BaseLLM.generate_prompt(self, prompts, stop, callbacks, **kwargs)
499 def generate_prompt(
500 self,
501 prompts: List[PromptValue],
(...)
504 **kwargs: Any,
505 ) -> LLMResult:
506 prompt_strings = [p.to_string() for p in prompts]
--> 507 return self.generate(prompt_strings, stop=stop, callbacks=callbacks, **kwargs)

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/base.py:656, in BaseLLM.generate(self, prompts, stop, callbacks, tags, metadata, run_name, **kwargs)
641 raise ValueError(
642 "Asked to cache, but no cache found at langchain.cache."
643 )
644 run_managers = [
645 callback_manager.on_llm_start(
646 dumpd(self),
(...)
654 )
655 ]
--> 656 output = self._generate_helper(
657 prompts, stop, run_managers, bool(new_arg_supported), **kwargs
658 )
659 return output
660 if len(missing_prompts) > 0:

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/base.py:544, in BaseLLM._generate_helper(self, prompts, stop, run_managers, new_arg_supported, **kwargs)
542 for run_manager in run_managers:
543 run_manager.on_llm_error(e)
--> 544 raise e
545 flattened_outputs = output.flatten()
546 for manager, flattened_output in zip(run_managers, flattened_outputs):

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/base.py:531, in BaseLLM._generate_helper(self, prompts, stop, run_managers, new_arg_supported, **kwargs)
521 def _generate_helper(
522 self,
523 prompts: List[str],
(...)
527 **kwargs: Any,
528 ) -> LLMResult:
529 try:
530 output = (
--> 531 self._generate(
532 prompts,
533 stop=stop,
534 # TODO: support multiple run managers
535 run_manager=run_managers[0] if run_managers else None,
536 **kwargs,
537 )
538 if new_arg_supported
539 else self._generate(prompts, stop=stop)
540 )
541 except BaseException as e:
542 for run_manager in run_managers:

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/openai.py:401, in BaseOpenAI._generate(self, prompts, stop, run_manager, **kwargs)
389 choices.append(
390 {
391 "text": generation.text,
(...)
398 }
399 )
400 else:
--> 401 response = completion_with_retry(
402 self, prompt=_prompts, run_manager=run_manager, **params
403 )
404 choices.extend(response["choices"])
405 update_token_usage(_keys, response, token_usage)

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/openai.py:109, in completion_with_retry(llm, run_manager, **kwargs)
103 def completion_with_retry(
104 llm: Union[BaseOpenAI, OpenAIChat],
105 run_manager: Optional[CallbackManagerForLLMRun] = None,
106 **kwargs: Any,
107 ) -> Any:
108 """Use tenacity to retry the completion call."""
--> 109 retry_decorator = _create_retry_decorator(llm, run_manager=run_manager)
111 @retry_decorator
112 def _completion_with_retry(**kwargs: Any) -> Any:
113 return llm.client.create(**kwargs)

File ~/miniconda3/envs/jupyter/lib/python3.11/site-packages/langchain/llms/openai.py:92, in _create_retry_decorator(llm, run_manager)
83 def _create_retry_decorator(
84 llm: Union[BaseOpenAI, OpenAIChat],
85 run_manager: Optional[
86 Union[AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun]
87 ] = None,
88 ) -> Callable[[Any], Any]:
89 import openai
91 errors = [
---> 92 openai.error.Timeout,
93 openai.error.APIError,
94 openai.error.APIConnectionError,
95 openai.error.RateLimitError,
96 openai.error.ServiceUnavailableError,
97 ]
98 return create_base_retry_decorator(
99 error_types=errors, max_retries=llm.max_retries, run_manager=run_manager
100 )

AttributeError: module 'openai' has no attribute 'error'

[BabellDev]

@dosuken123 I am trying your fix and it doesn't seem to work for minor openai versions such as 1.0.0b3. This comparison is returning False in that case:

I modified your fix to compare major version directly like so, and it seems to work:

def is_openai_v1() -> bool:
    _version = parse(version("openai"))
    return _version.major >= 1

EDIT: once I get past this check, there are other problems with the beta release of openai I'm using (1.0.0b3).

[efriis]

@baskaryan would love some eyes!

[efriis]

think this was mistakenly removed in a previous pr

[efriis]

OpenAIChat isn't actually compatible with v1. Just updated some easy things in case someone adds in future. Deprecated class

[efriis]

And thanks for getting this started @dosuken123 ! Added a few surrounding changes to make the compatibility match some of our other libs :)

[baskaryan]

didn't mean to approve, should remove httpx typing. otherwise LGTM!

[dosuken123]

Thanks @baskaryan and @efriis ! LangChain is awesome!

[adazoulay]

When can we expect the latest version of langchain to drop with these changes?

[efriis]

@adazoulay 0.0.334 is out!