Large Model Post-training & RL Scientist / Engineer

Large Model Post-training & RL Scientist / Engineer

Role Overview

You will work on post-training and reasoning for scientific foundation models, including SFT, RL, RLVR, alignment, scientific reasoning, and training-inference consistency. The goal is to make models more reliable, long-horizon, and verifiable on complex scientific problems.

Research Areas

• Post-training（SFT / DPO / RLHF / RLVR）

• Reinforcement Learning

• Large Model Alignment

• Scientific Reasoning

• Verifier / Reward Modeling

• Reinforcement Learning Infrastructure

• Training–Inference Consistency

• AI for Science

Responsibilities

• Post-training & Reinforcement Learning: Own end-to-end post-training for scientific foundation models, including SFT / DPO / RLHF / RLVR, and continuously improve scientific reasoning, instruction following, factuality, and complex problem-solving.

• Scientific Reasoning: Explore long-chain reasoning, verifiers, rewards, test-time compute, and related methods for scientific problems so models can solve more complex and verifiable tasks.

• Training Infrastructure: Ensure efficient and stable post-training and training-inference consistency while improving throughput, resource utilization, and experiment velocity.

• Data & Evaluation: Build high-quality reasoning / trajectory / preference / verifier data and evaluation systems for scientific reasoning and post-training, using results to drive model iteration.

What We Look For

• Bachelor’s degree or above in computer science, artificial intelligence, mathematics, or a related STEM field.

• Strong foundations in deep learning and large models, with hands-on experience in LLM post-training, reinforcement learning, or alignment and the ability to show verifiable results.

• Proficiency in PyTorch and at least one post-training / alignment framework such as veRL, TRL, OpenRLHF, or LLaMA-Factory.

• Understanding of SFT, DPO, RLHF, RLVR, and when to use each approach.

• Strong interest in frontier directions such as reasoning models, RL, verifiers, and test-time scaling.

• Long-term interest in AI for Science and the ability to independently drive post-training experiments and model iteration.

Nice to Have

• End-to-end project experience with SFT / DPO / RLHF / RLVR.

• Experience with reasoning models, verifiers, reward models, or test-time compute.

• Experience with post-training performance optimization, stability, or training-inference consistency.

• Experience building high-quality reasoning / trajectory datasets.

• Publications at strong conferences or journals.

• High-quality open-source contributions.

• Awards in AI-related competitions.