§ ÇPƒiª ãóJ—ddlmZGd„de¦«ZGd„de¦«ZdS)é)ÚProtocolcó:—eZdZdZd d„Zd d„Zd d„Zd d„Zd d„ZdS) ÚFTRecipeInterfacea This class provides a loose structure which every LLM fine-tuning recipe should follow. Please note that the interface itself should not be a vehicle for code reuse. torchtune strictly prohibits implementation inheritance in the codebase. A few notes about the design and the need for this interface: - This interface is meant to help recipe-writers organize their code in a way which is easy to read, understand and extend. Minimizing code duplication is not the goal. Recipe-writers are encouraged to copy-paste-modify. - This interface is not meant to add constraints. If the interface comes in the way of doing stuff, it needs to be updated or a new interface should be written to support what might be a new "family" of recipes. ÚreturnNc ó—dS)zÜ Responsible for loading ALL of the state for the recipe from the checkpoint file, including state for the model, optimizer, dataloader and training parameters such as the epoch and seed. N©©ÚselfÚkwargss úo/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/torchtune/recipe_interfaces.pyÚload_checkpointz!FTRecipeInterface.load_checkpointó €ð ˆóc ó—dS)z  Responsible for setting up all of the components necessary for training. This includes model, optimizer, loss function and dataloader. Nrr s r ÚsetupzFTRecipeInterface.setup"ó €ð ˆrc ó—dS)z„ All of the training logic, including the core loop, loss computation, gradient accumulation, and backward. Nrr s r ÚtrainzFTRecipeInterface.train)rrc ó—dS)z Responsible for saving ALL of the state for the recipe, including state for the model, optimizer, dataloader and training parameters such as the epoch and seed. Nrr s r Úsave_checkpointz!FTRecipeInterface.save_checkpoint0rrc ó—dS)z8 Any cleaning up needed for the recipe. Nrr s r ÚcleanupzFTRecipeInterface.cleanup8ó €ð ˆr©rN) Ú__name__Ú __module__Ú __qualname__Ú__doc__r rrrrrrr rr s‚€€€€€ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð rrcó*—eZdZdZdd„Zdd„Zdd„ZdS)ÚEvalRecipeInterfacezÿ This class provides a loose structure which every LLM evaluation recipe should follow. Please note that the interface itself should not be a vehicle for code reuse. torchtune strictly prohibits implementation inheritance in the codebase. rNc ó—dS)zk Responsible for loading ALL of the state for the recipe from the checkpoint file. Nrr s r r z#EvalRecipeInterface.load_checkpointFrrc ó—dS)z\ Responsible for setting up all of the components necessary for evaluation. Nrr s r rzEvalRecipeInterface.setupMrrc ó—dS)zC All of the evaluation logic, including reporting. Nrr s r ÚevaluatezEvalRecipeInterface.evaluateSrrr)rrrrr rr$rrr r r ?sZ€€€€€ððð  ð ð ð ð ð ð ð ð  ð ð ð ð ð rr N)Útypingrrr rrr úr&sxððÐÐÐÐÐð2 ð2 ð2 ð2 ð2 ˜ñ2 ô2 ð2 ðj ð ð ð ð ˜(ñ ô ð ð ð r