| Age | Commit message (Collapse) | Author | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
BREAKING CHANGE: rename model's `max_tokens` to `max_input_tokens`
|
|
|
|
* feat: compress session automaticlly
* non-block
* update field description
* set compress_threshold
* update session::clear_messages
* able to override session compress_threshold
* enable compress_threshold by default
* make session compress_threshold optional
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
* Add an "extra_fields" config to localai models
Because there are so many local AIs out there with a bunch of custom
parameters you can set, this allows users to send in extra parameters to
a local LLM runner, such as, e.g. `instruction_template: Alpaca`, so
that Mixtral can take a system prompt.
* support ollama
---------
Co-authored-by: sigoden <sigoden@gmail.com>
|
|
|
|
1. automatically switch to the model that has the necessary capabilities.
2. throw an error if the client does not have a model with the necessary capabilities
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|