Support Mistral (#133)

Support for gemma architecture models (#125)

added conversion to float32 before going to numpy (#127)

Added value error when insufficient training data is provided (#116)

Fixed the load artifact test (#130)

Fixed the load artifact test (#128)

Better logging in tests (#117)

Moved data shuffling to just before tokenization step (#110)

Better errors for when lens loading fails (#118)

Update loading_pretrained_lenses.rst (#120)

Generate from the lens (#109)

tests assert no warnings (#119)

Improved training docs (#115)

Fixed and simplified the CLI for the evaluation subcommand (#108)

Improve logging durring training and evaluation (#111)