Job Responsibilities
1. End-to-End Visual Language Action Model (VLA) Engineer
2. Design and optimize model architectures, perform data processing, training, and deployment on actual devices to ensure model efficiency and accuracy
3. Track cutting-edge advancements in VLA/VLM domains such as stream matching and action segmentation architectures, exploring novel model architectures and algorithm optimization techniques
Job Requirements
1. Possess professional experience in computer science, artificial intelligence, automation, robotics, or related fields, or hold a master's degree or higher in a relevant discipline.
2. Proficient in mainstream deep learning frameworks such as PyTorch, TensorFlow, and Jax, with strong programming skills.
3. Understanding of mainstream generative models (e.g., Diffusion, Flow) and familiarity with associated evaluation methods
4. Familiarity with transformer architectures, knowledge of end-to-end robotic operation models (e.g., ACT, OpenVLA, pi0, RT2), and cutting-edge advancements in vision-language-navigation large models
5. Experience training or fine-tuning VLA/VLM models
Bonus Points: Experience with model quantization and inference deployment on NVIDIA edge chips
Bonus Points: Preference given to candidates with ACM programming competition experience, or awards from top-tier conferences/competitions in natural language processing/multimodal domains
