Languagemodels

Changed

- Model names no longer include backend and quantization info
- Default to CPU inference unless GPU enabled using `lm.config["device"]="auto"`

Added

- Add quantization info to config and use it for memory usage calculation

Fixed

- Increase repetition penalty to 1.3 from 1.2 to help avoid repetition in smaller models

Changed

- Improve semantic meaning of chunk heading
- Remove sentencepiece dependency

Added

- Support GPT-based models
- Add `code` generation function
- Create new configuration system
- Use CUDA if available

Fixed

- Use non-greedy sampling on `complete` function
- Decrease chance of splitting chunks on decimal points
- Correct assistant example

Changed

- Attempt to chunk context on semantic boundaries

Added

- Allow filtering by model license

Fixed

- Update classification to only allow valid classes to be returned

Changed

- Disable beam search for faster inference

Changed

- Normalize output
- Rename some functions

Added

- Support xl models

Changed

- Less verbose chat syntax