v0.3

What's Changed

Framework

• Change: Now clem automatically assumes self-play when only a single model is given (and the game is multi-player) by @phisad
• Change: removed the option to run all games (as this simplifies the code and is now done in a pipeline script) by @phisad
• Change: Re-structured results folder (now grouped by model pairing instead of games) by @phisad in #33
• Added re-prompting hooks for DialogueGameMaster by @lpfennigschmidt in #42
• Add option to specify filename for instances in GameInstanceGenerator by @AnneBeyer in #44

CLI

• Change: Re-structured CLI calls (now have -g and -m option) by @phisad @briemadu in #35
• Add CLI option to set max_token to be generated by @phisad in #48

Backends

• Added Huggingface backend and documentation (which also introduces a hf local model registry) by @Gnurro in #2 #3 #17 #18 #29 #31 #32 #45 #46
• Added Cohere API by @sherzod-hakimov
• Added Mistral API by @sherzod-hakimov
• Added generic openai compatible API by @davidschlangen

Documentation

• Improved docs on how to run the benchmark by @davidschlangen @sherzod-hakimov @briemadu
• Improved docs on how to add and prototype games by @phisad @briemadu @davidschlangen
• Improved docs on how to log games by @briemadu
• Improved docs on how to add a backend by @briemadu #21
• Improved docs on how to run games locally by @sherzod-hakimov @AnneBeyer

Dependencies

• bumped accelerate to 0.25.0
• bumped transformers to 4.36.0
• bumped openai to 1.7.0
• bumped anthropic to 0.3.0
• added cohere 4.34
• added mistralai 0.0.12

... and other small changes.

Full Changelog: v0.2...v0.3

Version 0.2

Submission version