UnprocessableEntityError when embedding using openai and a proxy

[MoezGholami]

Description

Original issue on OpenAI community forum: https://community.openai.com/t/attributeerror-module-openai-has-no-attribute-error/486676

The new version of the openai package is breaking the the Jupyter-ai notebook. Any command to the openai models raises the following error:

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

The Jupyter notebook works with openai version 0.28 but fails with openai 1.2.3.

Reproduce

1. pip install --upgrade openai
2. Send the any command to Open AI's API in a Juypter notebook.

Expected behavior

Jupyter AI should work with the new openai python package.

Context

Versions:

1. jupyter_ai: 2.5.0
2. openai: 1.2.3
3. python: 3.11.6
4. pip: 23.3.1

[welcome]

Thank you for opening your first issue in this project! Engagement like this is essential for open source projects! 🤗

If you haven't done so already, check out Jupyter's Code of Conduct. Also, please try to follow the issue template as it helps other other community members to contribute more effectively.

You can meet the other Jovyans by joining our Discourse forum. There is also an intro thread there where you can stop by and say Hi! 👋

Welcome to the Jupyter community! 🎉

[dlqqq]

Quick dig through the LangChain PRs indicates that this was fixed last week, first released at v0.0.336. Should be fixable by bumping our LangChain version.

Edit: forgot to include link. langchain-ai/langchain#13262

[sqlreport]

@dlqqq I have:
python=3.10
jupyter_ai==1.9.0
langchain==0.0.350
openai==1.7.1

when I run /learn command on chatUI, I get an error
IndexError: list index out of range

[sqlreport]

When I change my documents to txt file I am getting error like (looks like expecting string input instead)

Exception: UnprocessableEntityError("Error code: 422 - {'detail': [{'type': 'string_type', 'loc': ['body', 'input', 'str'], 'msg': 'Input should be a valid string', 'input': [[2465, 836, 374, 350, 6043, 12336]], 'url': 'https://errors.pydantic.dev/2.4/v/string_type'}, {'type': 'string_type', 'loc': ['body', 'input', 'list[str]', 0], 'msg': 'Input should be a valid string', 'input': [2465, 836, 374, 350, 6043, 12336], 'url': 'https://errors.pydantic.dev/2.4/v/string_type'}]}")

[JasonWeill]

@sqlreport This is likely a regression after PR #551 was merged in and released in version 1.9.0 and 2.9.0. I can look at it.

[JasonWeill]

With tip-of-main in Jupyter AI and openai 1.6.1, I don't see this error with an OpenAI embedding model; /learn on a directory including text files works without errors.

[sqlreport]

@JasonWeill can you provide unit test code that can mimic the openai call so I can troubleshoot my openai proxy? I have ran openai example code with string and it works without issue. It must be the way chatUI generates the openai call where it is passing non-string data.

[JasonWeill]

@sqlreport I looked at our existing codebase and I don't see unit tests for the /learn handler, sorry. There are two possible places where unexpected non-string data might come in:

1. The split method in packages/jupyter-ai/jupyter_ai/document_loaders/directory.py:

   jupyter-ai/packages/jupyter-ai/jupyter_ai/document_loaders/directory.py

   Lines 51 to 71 in 814eb44

   	def split(path, all_files: bool, splitter):
   	chunks = []
   	for dir, subdirs, filenames in os.walk(path):
   	# Filter out hidden filenames, hidden directories, and excluded directories,
   	# unless "all files" are requested
   	if not all_files:
   	subdirs[:] = [d for d in subdirs if not (d[0] == "." or d in EXCLUDE_DIRS)]
   	filenames = [f for f in filenames if not f[0] == "."]
   	for filename in filenames:
   	filepath = Path(os.path.join(dir, filename))
   	if filepath.suffix not in SUPPORTED_EXTS:
   	continue
   	document = dask.delayed(path_to_doc)(filepath)
   	chunk = dask.delayed(split_document)(document, splitter)
   	chunks.append(chunk)
   	flattened_chunks = dask.delayed(flatten)(*chunks)
   	return flattened_chunks

   — this takes a file and converts it to chunks, according to one of several splitter classes.
2. The get_embeddings method in the same file:

   jupyter-ai/packages/jupyter-ai/jupyter_ai/document_loaders/directory.py

   Lines 95 to 104 in 814eb44

   	def get_embeddings(chunks, em_provider_cls, em_provider_args):
   	# split documents in parallel w.r.t. each file
   	embeddings = []
   	# compute embeddings in parallel
   	for chunk in chunks:
   	embedding = dask.delayed(embed_chunk)(chunk, em_provider_cls, em_provider_args)
   	embeddings.append(embedding)
   	return dask.delayed(join)(embeddings)

   — this creates a list of Dask delayed tasks to send one chunk at a time to the selected embedding model class (em_provider_cls). Dask executes these in learn.py, the /learn handler:

   jupyter-ai/packages/jupyter-ai/jupyter_ai/chat_handlers/learn.py

   Line 161 in 814eb44

   embedding_records = await dask_client.compute(delayed)

[andrewbovey]

@sqlreport Did you find a solution to the Exception: UnprocessableEntityError("Error code: 422 - {'detail': [{'type': 'string_type', 'loc': ['body', 'input', 'str'], 'msg': 'Input should be a valid string', 'input': [[2465, 836, 374, 350, 6043, 12336]], 'url': 'https://errors.pydantic.dev/2.4/v/string_type'}, {'type': 'string_type', 'loc': ['body', 'input', 'list[str]', 0], 'msg': 'Input should be a valid string', 'input': [2465, 836, 374, 350, 6043, 12336], 'url': 'https://errors.pydantic.dev/2.4/v/string_type'}]}")

Mine is UnprocessableEntityError: Error code: 422 - {'detail': [{'type': 'string_type', 'loc': ['body', 'input', 'str'], 'msg': 'Input should be a valid string', 'input': [[3923, 14071, 3956, 527, 1070, 30]]}, {'type': 'string_type', 'loc': ['body', 'input', 'list[str]', 0], 'msg': 'Input should be a valid string', 'input': [3923, 14071, 3956, 527, 1070, 30]}]}

langchain==0.2.1
opeanai==1.31.0