Tim Dettmers Portfolio
Assistant Professor at Carnegie Mellon University (CMU) and Research Scientist at the Allen Institute for Artificial Intelligence (Ai2)
Assistant Professor at Carnegie Mellon and Research Scientist at Ai2; first author of LLM.int8() and QLoRA and creator of bitsandbytes, the library that ships them, cutting the GPU memory needed to run and fine-tune large language models; his blog pairs posts on quantization and hardware with a GPU-buying guide carrying 1,665 comments, and since late 2025 essays on coding agents.
Location
Not listed — probably too busy shipping to update the bio.
Companies
Credentials
- PhD, University of Washington (advised by Luke Zettlemoyer)
Skills
Notable
- First author of QLoRA (NeurIPS 2023 oral) and LLM.int8() (NeurIPS 2022)
- Creator and maintainer of bitsandbytes (8.5k GitHub stars)
- 2025 Google ML and Systems Junior Faculty Award; 2024 AI2050 Early Career Fellow
- 2023 Google Open Source Award and PyTorch Foundation Award (per his About page)
What makes it work
A breakdown of the choices that make this portfolio stand out.
Research that ships as usable code, not just a paper
LLM.int8() and QLoRA ship through bitsandbytes, the library he created and still maintains (8.5k GitHub stars), so most people meet the research as a pip install rather than a PDF.
Technical writing that also covers the practical hardware layer
The GPU guide treats hardware economics as part of doing ML research, not a separate topic, which is unusually honest about what actually limits most people's ability to work with large models.
A comment section treated as part of the content
1,665 comments on the GPU guide alone show the writing functions as a living community reference rather than a one-way broadcast.
What AI Researchers can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
Ship your research as runnable code whenever possible — a technique people can actually install gets cited and used far more than a paper alone.
Write about the practical constraints (hardware, cost) around your research area, not just the algorithms — it's often the more-read content.
Let comments and community discussion accumulate on your best posts instead of routing everything to social media — a good comment section becomes part of the reference itself.
Projects
Tim Dettmers publishes a quantization method, then the library that runs it, then the guide to the hardware it runs on, which covers most of what stands between a reader and a working large model.
Fine-tunes a 4-bit quantized language model through low-rank adapters, so large models can be fine-tuned on far less GPU memory.
Research that changed what practitioners could do on the hardware they already had.
First-author paper
Implementation shipped through bitsandbytes
Oral presentation at NeurIPS 2023
The library behind 8-bit and 4-bit quantization and 8-bit optimizers in PyTorch, where LLM.int8() and QLoRA reach users.
Shows the rare pairing of new research methods with production-grade software that ships them.
Created and maintained by him
Moved to a foundation organization
The software home of his papers
Which GPU(s) to Get for Deep Learning
A long-running guide on choosing GPUs for deep learning: memory, tensor cores, caches, and cost-efficiency, last revised January 2023.
Treats hardware economics as part of doing ML research, not an afterthought.
Tagged across AMD, CPU, PCIe and sparse training topics
Explains why, not only what to buy
Comment thread works as a reader Q&A
More AI Researcher portfolios
Ready to build your portfolio?
Follow the AI Researcher roadmap — skills, projects, and timeline to get hired.