Your new company

Your new role

• Advancing Mixture of Experts (MoE) through dynamic routing, sparse activation, and scalable training architectures.
• Innovating Transformer-MoE integration to enhance model efficiency and performance in large-scale deployments.
• Optimizing Multi-Head Attention (MHA) via sparse and linear mechanisms, with theoretical and multimodal enhancements.
• Pioneering tokenization research, including subword/byte-level algorithms and multilingual framework development.
• Enhancing embedding techniques through compression and semantic space optimization for improved model representation.

What you'll need to succeed

• Holds a PhD in Computer Science, AI, Mathematics, or a closely related discipline.
• Demonstrates strong research experience in MoE, MHA, or Tokenization, with publications or deep project work.
• Skilled in PyTorch/JAX and experienced in training and fine-tuning large-scale models.
• Deeply familiar with Transformer architectures and frameworks like Megatron and DeepSpeed.
• Combines academic curiosity with an engineering mindset and a self-driven approach to practical implementation.

What you'll get in return

• Structured career growth and plenty of developmental opportunities
• Stable working environment

What you need to do now