Log exceptions in /generate to a file

[dlqqq]

• Fixes Save log locally when /generate command fails #422.

Description

• Performs two method renames:
  • BaseChatHandler.process_message() => BaseChatHandler.on_message()
  • BaseChatHandler._process_message() => BaseChatHandler.process_message()
  • While the old syntax was technically correct in how it uses public and private methods, I found this to be needlessly obscure. So I renamed the methods and added doc comments to clarify the difference between these two methods.
• Adds a new method on BaseChatHandler for chat handlers to override: self.handle_exc(). This allows chat handlers to define their own way of handling exceptions. If this is undefined, a default implementation is provided by BaseChatHandler.
• Logs exceptions raised by GenerateChatHandler to a timestamped file under ./jupyter-ai-logs/.

Demo

[ellisonbg]

I think this is a good approach as a fallback, but given that a common failure mode for /generate is that a model doesn't produce valid JSON, we should catch that error and handle it in a more user centered manner. Otherwise, users are going to look at that log file and open a GitHub issue this repo, when really we should be informing them that the model they were using wasn't able to generate valid JSON for the prompt and that they should try a different model.

[dlqqq]

@ellisonbg I am tracking your suggestion as a separate issue in #435. Making this change will take more time as we need to reliably detect when an LLM is failing to generate JSON, and distinguish that from other exceptions. Currently, I am seeing a double exception emitted from LangChain when I use some less-capable language models:

Traceback (most recent call last):
  File "/Users/dlq/miniconda3/envs/jai/lib/python3.11/site-packages/langchain/output_parsers/pydantic.py", line 27, in parse
    json_object = json.loads(json_str, strict=False)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/dlq/miniconda3/envs/jai/lib/python3.11/json/__init__.py", line 359, in loads
    return cls(**kw).decode(s)
           ^^^^^^^^^^^^^^^^^^^
  File "/Users/dlq/miniconda3/envs/jai/lib/python3.11/json/decoder.py", line 337, in decode
    obj, end = self.raw_decode(s, idx=_w(s, 0).end())
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/dlq/miniconda3/envs/jai/lib/python3.11/json/decoder.py", line 353, in raw_decode
    obj, end = self.scan_once(s, idx)
               ^^^^^^^^^^^^^^^^^^^^^^
json.decoder.JSONDecodeError: Expecting ':' delimiter: line 20 column 28 (char 436)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/Volumes/workplace/jupyter-ai/packages/jupyter-ai/jupyter_ai/chat_handlers/base.py", line 43, in on_message
    await self.process_message(message)
  File "/Volumes/workplace/jupyter-ai/packages/jupyter-ai/jupyter_ai/chat_handlers/generate.py", line 262, in process_message
    final_path = await self._generate_notebook(prompt=message.body)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Volumes/workplace/jupyter-ai/packages/jupyter-ai/jupyter_ai/chat_handlers/generate.py", line 238, in _generate_notebook
    outline = await generate_outline(prompt, llm=self.llm, verbose=True)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Volumes/workplace/jupyter-ai/packages/jupyter-ai/jupyter_ai/chat_handlers/generate.py", line 56, in generate_outline
    outline = parser.parse(outline)
              ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/dlq/miniconda3/envs/jai/lib/python3.11/site-packages/langchain/output_parsers/pydantic.py", line 33, in parse
    raise OutputParserException(msg, llm_output=text)
langchain.schema.output_parser.OutputParserException: Failed to parse Outline from completion