Responsibilities

1. Pre-training development and tuning of large language models and multimodal large models.

2. Design, training, dataset processing, and real-robot deployment testing of large-model-based robot control algorithms.

3. Improving large-model training efficiency (e.g., model parallelism, flash attention, LoRA).

4. Tracking the latest advances and pre-researching and evaluating large-model applications in robotics (e.g., the RT series).

Requirements

1. Solid grasp of network structures and training methods for large language and multimodal models.

2. Proficient in PyTorch/TensorFlow and at least one of Python/C++.

3. Proficient in common distributed-training frameworks and efficient fine-tuning of large models.

4. Skilled in using cloud computing power for large-scale data processing and training.

5. Proficiency in common imitation-learning algorithms (e.g., ACT, DP) and Model-Based RL is a plus.

6. Experience wiring model-to-robot hardware interfaces is a plus.

7. Papers in top AI/CV/NLP venues (e.g., TIP, TRO, CVPR, ACL) are a plus.