Senior Researcher – Diffusion-Based Large Language Models

About the role

Huawei Canada’s Noah’s Ark Lab is seeking a Senior Researcher to lead cutting‑edge work on diffusion‑based large language models. The role focuses on advancing generative text and multimodal modeling, bridging theory and large‑scale production systems.

Key responsibilities

• Propose and develop novel diffusion‑style architectures for LLMs, including discrete and continuous diffusion, denoising objectives, and hybrid transformer‑diffusion models.
• Design and run large‑scale experiments, implement training pipelines, and analyse results to improve sample quality, controllability, and inference efficiency.
• Collaborate with data and infrastructure teams on data curation, evaluation, fine‑tuning, RLHF, safety and robustness of diffusion LLMs.
• Publish research in top‑tier ML/NLP venues and contribute to open‑source libraries or model releases.
• If senior, supervise and mentor a small group of researchers and engineers.

Required profile

• Ph.D. in Computer Science, Machine Learning, Statistics, Applied Mathematics or equivalent research experience.
• First‑author publications in top venues such as NeurIPS, ICML, ICLR, ACL, EMNLP or NAACL.
• Deep understanding of generative modeling, diffusion models and their relationship to autoregressive LLMs.
• Strong theoretical grounding in optimization, probabilistic modeling and representation learning.

Required skills

• Python programming with extensive PyTorch experience.
• Proficiency with Hugging Face Transformers, Diffusers, Accelerate and large‑scale training frameworks (DeepSpeed, Megatron‑LM, vLLM).
• Experience building distributed training pipelines and large‑scale experimentation.
• Hands‑on knowledge of modern Transformer‑based LLMs (e.g., Llama, Qwen) and diffusion/score‑based models for text, image or multimodal generation.
• Ability to evaluate generative models using both automatic metrics and human assessment.