Explore-Execute-Chain

Explore–Execute Chain (E2C) Model

This repository hosts the pretrained and fine-tuned Explore–Execute Chain (E2C) models.

Paper: Explore–Execute Chain: Towards an Efficient Structured Reasoning Paradigm
Kaisen Yang, Lixuan He, Rushi Shah, Kaicheng Yang, Qinwei Ma, Dianbo Liu, Alex Lamb

Under review at ICLR 2026

Code: GitHub – Explore–Execute Chain

---

📖 Model Overview

E2C is a two-stage reasoning framework designed to improve the efficiency and interpretability of large language models (LLMs):

1. Exploration — Generate lightweight reasoning sketches (plans).
2. Execution — Execute selected plans faithfully for high-quality results.

Benefits:

• Efficient reasoning with minimal computation
• Explicit, interpretable exploration traces
• Easy domain adaptation with minimal supervision

---

🚀 Key Features

• Two-stage training
  • E2C-SFT — Supervised fine-tuning on exploration–execution pairs
  • E2C-RL — Reinforcement learning to refine execution
• Efficient adaptation (EF-SFT) — Adapt with exploration-only data
• Test-time scaling — Aggregate multiple explorations for better results
• Benchmarked on mathematical and medical reasoning datasets

---

📥 Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "KaisenYang/Explore-Execute-Chain"
model_type = "8B-Final"  # change to the subfolder you want to use

tokenizer = AutoTokenizer.from_pretrained(model_name, subfolder=model_type)
model = AutoModelForCausalLM.from_pretrained(model_name, subfolder=model_type)

# Test example: Fibonacci sequence
inputs = tokenizer("What is the 10th number in the Fibonacci sequence?", return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0]))

---

🔗 Links

• 📂 Full code and experiments: GitHub Repository
• 📜 Paper (under review): ICLR 2026 submission

---

📜 Citation

If you use this work, please cite:

@inproceedings{yang2026explore,
  title={Explore-Execute Chain: Towards an Efficient Structured Reasoning Paradigm},
  author={Yang, Kaisen and He, Lixuan and Shah, Rushi and Yang, Kaicheng and Ma, Qinwei and Liu, Dianbo and Lamb, Alex},
  booktitle={International Conference on Learning Representations (ICLR)},
  year={2026},
  note={under review}
}

---

🧾 License

This project is licensed under the MIT License. See the LICENSE file for details.