saurabh-singh-rajput/green-tea-llama-3.1-8b-grpo

Llama-3.1-8B GRPO

Part of the Green Tea replication package for Beyond the Need for Speed: Energy-Aware Code Generation via Simulation-Guided Reinforcement Learning (Rajput and Sharma). Simulation-in-the-loop GRPO on the Energy-SFT model. On the 143-problem held-out benchmark it is the cross-family closed-loop GRPO checkpoint.

• Base model: saurabh-singh-rajput/green-tea-llama-3.1-8b-energy-sft
• Training: Simulation-in-the-loop GRPO on the Energy-SFT model
• Code and full replication: https://github.com/SMART-Dal/green-tea
• Dataset (Zenodo): https://doi.org/10.5281/zenodo.21210099

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
b = AutoModelForCausalLM.from_pretrained('saurabh-singh-rajput/green-tea-llama-3.1-8b-energy-sft', device_map='auto')
m = PeftModel.from_pretrained(b, 'saurabh-singh-rajput/green-tea-llama-3.1-8b-grpo')
t = AutoTokenizer.from_pretrained('saurabh-singh-rajput/green-tea-llama-3.1-8b-grpo')

Citation

@misc{rajput2026greentea,
  title={Beyond the Need for Speed: Energy-Aware Code Generation via Simulation-Guided Reinforcement Learning},
  author={Rajput, Saurabhsingh and Sharma, Tushar},
  year={2026},
  note={Preprint}
}