Faculty Profile: Prof. Tim Dettmers — CMU MLD + CSD
Faculty Profile: Prof. Tim Dettmers — CMU MLD + CSD
Position: Assistant Professor (Machine Learning Dept + Computer Science Dept, joint) Institution: Carnegie Mellon University Report date: 2026-06-12
Research Focus
LLM quantization, parameter-efficient fine-tuning (PEFT), on-device and accessible AI, LLM efficiency/accessibility, agent systems for non-AI experts.
Academic Profile
- PhD: University of Amsterdam
- QLoRA (2023): NeurIPS — most cited paper on efficient LLM fine-tuning; adopted widely in Llama, Mistral, etc.
- Google ML and Systems Junior Faculty Award
- AI2050 Early Career Fellow (2025)
- AI2050 project: develop agent systems to help non-AI experts adapt AI models to specific domains (medical sciences, etc.)
Key Publications
| Paper | Venue | Focus |
|---|---|---|
| QLoRA: Efficient Finetuning of Quantized LLMs | NeurIPS’23 | 4-bit quantization + LoRA; 65B model fine-tuned on 1 GPU; widely adopted |
| bitsandbytes library | Ongoing | GPU-efficient quantization for LLMs |
| Sparse fine-tuning scaling | 2024 | Parameter-efficient methods for large models |
Fit with Weijia Zhang
| Dimension | Assessment |
|---|---|
| Efficient SFT (QLoRA, PEFT) | ✅ World-class expertise |
| LLM accessibility / on-device | ✅ Core focus |
| Agent systems for domain experts | ✅ AI2050 project direction |
| RL post-training | ❌ Not his focus |
| NLP agents / agentic AI | ⚠️ Adjacent via AI2050 project |
| GUI / VLM agents | ❌ Not his area |
Verdict
P3 套磁（若做 efficient SFT / on-device agent 方向）。 QLoRA is directly relevant to Weijia’s SFT work at MSRA. If Weijia’s research involves efficient fine-tuning or deploying agents on limited compute, Dettmers is the go-to expert. Less relevant for pure agentic AI capability work.