Open JasonWeill opened 2 months ago
To avoid breaking changes with deleting these aliases, I think a nice feature would be for users to specify aliases in the startup configuration.
This would allow users to specify these aliases (amongst others they might always use) in the configuration instead of running the %ai register
command in their notebooks. And those aliases will be available without the need for re-registering them in a notebook.
It is possible to add traitlet configuration to magics, in IPython core this is done for %store
magic here and then user just sets c.StoreMagics.autorestore = True
in their config file. I think these aliases should be moved to a Dict traitlet.
Problem
A fresh installation of Jupyter AI has the following aliases predefined. Several of these are obsolete or little-used, and by pre-defining these aliases, we developers seem to be endorsing these models.
gpt2
huggingface_hub:gpt2
gpt3
openai:davinci-002
chatgpt
openai-chat:gpt-3.5-turbo
gpt4
openai-chat:gpt-4
ernie-bot
qianfan:ERNIE-Bot
ernie-bot-4
qianfan:ERNIE-Bot-4
titan
bedrock:amazon.titan-tg1-large
Proposed Solution
Remove these aliases from the default Jupyter AI configuration. This may be a breaking change, if users rely on these aliases.
Additional context
Users could not add custom model aliases when Jupyter AI first launched (#45) but this was added with #136, in June 2023.