interpretml / talktoebm Goto Github PK

Hi, I wanted to try out your interesting project locally but I a currently receive the error message

"raise Exception(f"Too many (more than {self.llm.max_retries}) OpenAI API RateLimitError's in a row!")
Exception: Too many (more than 5) OpenAI API RateLimitError's in a row!"

when calling the "llm_describe_ebm_graph(...)" function.

It looks like too many requests were sent to the API within a short period of time, more than the API allows.
I suppose that this problem did not occur when you made your repo public 8 months ago but rather occurs due to changes in newer guidance and openai packages.
I did indeed find out that your t2ebm package only seems to work with openai package versions >= 0.27.10 and <= 0.28.1.
Because when using a newer version I receive the following error message:

"File "[...]\talk2ebm\Lib\site-packages\t2ebm\utils.py", line 24, in \t2ebm\utils.py", line 24, in
retry=retry_if_not_exception_type(openai.error.InvalidRequestError),
^^^^^^^^^^^^
AttributeError: module 'openai' has no attribute 'error'"

At first I thought that this error (the RateLimitError on top) might occur due to the complexity of my dataset so I tried calling the function on the Spaceship Titanic dataset (the one you used in your demo notebook) but the error still occurs. I also tried calling the function with a different api_key from another Openai account but the error still occurs.

I would be very nice, if you could try to replicate the error to confirm that this is in fact an issue based on an outdated "llm_describe_ebm_graph(...)" function.

I encountered an issue when trying to run the following code in the TalkToEBM package:

llm_descripe_dict = {k: kwargs[k] for k in dict(kwargs) if k in llm_descripe_kwargs}
prompt = prompts.describe_graph_cot(
graph, num_sentences=num_sentences, **llm_descripe_dict
)

return guidance(prompt, llm, silent=True)()["cot_graph_description"]

Error Message:
TypeError: _Guidance.call() got an unexpected keyword argument 'silent'

Steps to Reproduce:
Install the TalkToEBM package from GitHub.

Run the code snippet:
graph_description = t2ebm.llm_describe_ebm_graph(llm, ebm, 0) # feature 0, 'HomePlanet'
print(textwrap.fill(graph_description, 80))

Expected Behavior:
The code should execute without errors and return the description of the graph.

Actual Behavior:
The code raises a TypeError indicating that the silent keyword argument is unexpected in the _Guidance.call() method.

Environment:
TalkToEBM version: latest from GitHub
Operating System: Windows 10

Please let me know if there is a workaround for this issue or if I need to update any dependencies.
Thank you!