isotropic-labs/last_mlp_seed7

This repository contains the last_mlp_seed7 head-diverse reward model checkpoint from mc-dropout-vs-reward-ensemble.

Loading

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model_name = "isotropic-labs/last_mlp_seed7"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForSequenceClassification.from_pretrained(
    model_name,
    trust_remote_code=True,
)

These checkpoints use a custom GPT-NeoX reward-model class, so trust_remote_code=True is required unless you register the class locally from this repository.