def load_llama_models_tokenizer()

in evaluator/evaluator.py [0:0]

• 8 lines of code
• 1 McCabe index (conditional complexity)

def load_llama_models_tokenizer(args):
    tokenizer = Llama2Tokenizer.from_pretrained(args.checkpoint_path)
    model = Model.from_pretrained(
        args.checkpoint_path,
        device_map=f'cuda:{args.gpu}')
    model.generation_config.do_sample = False  # use greedy decoding
    model.generation_config.repetition_penalty = 1.0  # disable repetition penalty
    return model, tokenizer