sms1097/utility_model

Utility Model

This generates the IsUseful token as descirbed in Self-RAG.

We are testing to see if an answer is useful to the given user question. We output a score from 1-5 based on how useful the answer is.

The expected input to the model is:

Instruction: {instruction}\nAnswer: {answer}",

Training Results

{'eval_loss': 0.4719298779964447, 
'eval_mse': 0.4719298183917999,
'eval_mae': 0.25655537843704224,
'eval_r2': 0.5200293292355334,
'eval_accuracy': 0.9001516683518705}