Weaver Distilled for GPQA

This is a distilled cross-encoder model based on ModernBERT-large, trained to predict the correctness of answers on GPQA. This specialized verifier was trained on Weaver scores aggregated over 35 different verifiers and reward models.

Model Details

• Base Model: answerdotai/ModernBERT-large (395M parameters)
• Architecture: Cross-encoder with MLP head (1024 → 512 → 256 → 1)
• Max Sequence Length: 4096 tokens
• Training Data: GPQA problems with Weaver scores from 35 LM judges and reward models
• Task: Binary classification for answer correctness prediction

Quick Start

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load model and tokenizer
model_name = "hazyresearch/Weaver_Distilled_for_GPQA"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Example usage
instruction = "Which of the following best describes photosynthesis? A) Converting light to chemical energy B) Breaking down glucose C) Cellular respiration D) Protein synthesis"
response = "The answer is A) Converting light to chemical energy. Photosynthesis is the process by which plants convert light energy into chemical energy..."

# Tokenize input pair
inputs = tokenizer(
    instruction, 
    response,
    truncation=True,
    max_length=4096,
    padding=True,
    return_tensors="pt"
)

# Get correctness score
with torch.no_grad():
    outputs = model(**inputs)
    score = torch.sigmoid(outputs.logits).item()
    
print(f"Correctness score: {score:.3f}")
print(f"Prediction: {'Correct' if score > 0.5 else 'Incorrect'}")

Citation

@misc{saadfalcon2025shrinkinggenerationverificationgapweak,
      title={Shrinking the Generation-Verification Gap with Weak Verifiers}, 
      author={Jon Saad-Falcon and E. Kelly Buchanan and Mayee F. Chen and Tzu-Heng Huang and Brendan McLaughlin and Tanvir Bhathal and Shang Zhu and Ben Athiwaratkun and Frederic Sala and Scott Linderman and Azalia Mirhoseini and Christopher Ré},
      year={2025},
      eprint={2506.18203},
      archivePrefix={arXiv},
      primaryClass={cs.CR},
      url={https://arxiv.org/abs/2506.18203}, 
}