Team Introduction
The Doubao (Seed) Vision AI Platform team focuses on the end-to-end infrastructure development and efficiency improvement for Seed vision-based large model development, including the data pipeline construction and training, evaluation data delivery, and full lifecycle efficiency enhancement for visual large models such as VLM, VGFM, and T2I. This also encompasses large-scale training stability and optimization for acceleration, as well as large model inference and multi-machine multi-card deployment.
We are looking for talented individuals to join us for a Student Researcher opportunity in 2025. Student Researcher opportunities at ByteDance aim to offer students industry exposure and hands-on experience. Turn your ambitions into reality as your inspiration brings infinite opportunities at ByteDance.
The Student Researcher position provides unique opportunities that go beyond the constraints of our standard internship program, allowing for flexibility in duration, time commitment, and location of work.
Candidates can apply to a maximum of two positions and will be considered for jobs in the order you apply. The application limit is applicable to ByteDance and its affiliates' jobs globally. Applications will be reviewed on a rolling basis - we encourage you to apply early.
Responsibilities:
1. Develop algorithm acceleration technologies for text-to-image/text-to-video models through knowledge distillation, model architecture redesign (dynamic MoE routing/sparse attention), and parameter-efficient design (low-bit quantization) to achieve order-of-magnitude efficiency gains.
2. Lead generative model innovation with focus on diffusion acceleration (sampling step reduction/latent optimization), autoregression model efficiency.
3. Collaborate cross-functionally to identify performance bottlenecks, optimize vision models via algorithmic breakthroughs, and enhance ByteDance's product capabilities.
Minimum Qualifications
1. Currently pursuing a PhD in Software Development, Computer Science, Computer Engineering, or a related technical discipline.
2. Expertise in diffusion models (Stable Diffusion/DiT) with deep understanding of computational bottlenecks and optimization methodologies.
3. Proven experience in ≥1 domain: model compression (quantization/knowledge distillation), efficient architectures (MoE/sparse attention), generative alignment (RLHF/DPO);
4. Must obtain work authorization in country of employment at the time of hire, and maintain ongoing work authorization during employment.
Preferred Qualification
1. Kaggle competition achievements, publications at ICML/NeurIPS/CVPR, or open-source contributions (e.g., HuggingFace Diffusers optimization).
Read Full Description