[Bug] load_model not using revision

[joshlevy89]

In the examples (and evals on honesty-eval branch), when load_model is called in the setup of the notebook, the "revision" parameter is only passed to the model from_pretrained, but not the tokenizer from_pretrained. I doubt this makes a difference - probably the tokenizer is the same for the revision as for the main branch for the models I've used - but nevertheless, should be added.

Should add what is in bold.

def load_model(model_name_or_path, revision, device):
model = AutoModelForCausalLM.from_pretrained(
model_name_or_path, device_map=device, revision=revision, trust_remote_code=False)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True, padding_side="left", revision=revision)
tokenizer.pad_token_id = 0
return model, tokenizer