Introduces several changes with the main objective of enabling OpenAI's streaming API in the chatmastermind application. This allows for the retrieval of AI responses gradually as a stream, which can significantly improve the user experience in interactions that involve large result sets.
* Added tiktoken import in 'openai.py' and modifications to the OpenAI class to support streaming. This includes the addition of a new class OpenAIAnswer to handle streaming API responses.
* Modified request function in the OpenAI class: the stream=True flag is added to the openai.ChatCompletion.create method to enable streaming API.
* Modified 'question.py' to print the answer parts as they are streamed.
* Replaced the Answer class's string data type with a generator which supports str and Generator[str, None, None] data types. Modifications are made to the Answer class methods to handle both data types accordingly.
* Updated the tests in 'test_ais_openai.py' and 'test_message.py' to reflect and validate these changes.
- no longer using file suffix to choose the format
- added 'mformat' argument to 'write_xxx()' functions
- file suffix is now set by 'Message.to_file()' per default
- '.msg' suffix is always used for writing
- 'Message.to_file()' will set the file suffix if the given file_path has none
- added 'mformat' argument to 'Message.to_file()' for choosing the file format
- '.txt' and '.yaml' suffixes are only supported for reading
This PR primarily modifies the `cmm question --repeat` command to allow repeating multiple questions, instead of only the last one.
Additionally, this PR includes the following changes:
- In `ai_factory.py`, added optional parameters 'def_ai' and 'def_model' to the `create_ai` function which allows specifying a default AI and model.
- In `openai.py`, a potential bug was fixed where the 'tags' attribute was updated to ensure it is always a set, even when 'otags' is None.
- In `question.py`, a significant amount of new code was added to facilitate the 'repeat' functionality. This includes functions to create modified args based on an existing message (`create_msg_args`), to repeat a given list of messages (`repeat_messages`), and to invert the semantics of the INPUT tags for this command (`invert_input_tag_args`).
- In `main.py`, the 'nargs' parameter was changed from `+` to `*` in the 'or-tags', 'and-tags', and 'exclude-tags' arguments to accommodate the updated handling of tags in `question.py`.
- A new `test_common.py` file was added which includes a `FakeAI` class for testing purposes, and a `TestWithFakeAI` class which includes a number of methods for asserting various conditions about messages.
This PR also includes additional tests to verify the correct operation of the new 'repeat' functionality.
