|
Saleh Ashkboos
AI Solution Architect at NVIDIA
Zurich, Switzerland. |
I'm an AI Solution Architect at NVIDIA. I am focused on accelerating deep neural network training using quantization. Before joining NVIDIA, I received my Ph.D. in Computer Science from ETH Zurich, advised by Professor Torsten Hoefler and Professor Dan Alistarh.
2026/05: I joined NVIDIA in Zurich.
2026/01: Two papers are accepted at ICLR26: "Beyond Outliers: A Study of Optimizers Under Quantization" and "Bridging the Gap Between Promise and Performance for Microscaling FP4 Quantization".
2025/11: I passed my Ph.D. defense! Thanks to my advisor Prof. Torsten Hoefler, and my committee members: Prof. Dan Alistarh and Dr. Mahyar Najibi.
2025/10: New preprint is out: "Beyond Outliers: A Study of Optimizers Under Quantization".
2025/01: New preprint is out: "HALO: Hadamard-Assisted Lossless Optimization for Efficient Low-Precision LLM Training and Fine-Tuning".
2024/09: "QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs " accepted at NeurIPS24. (Github)
2024/09: "QUIK: Towards End-to-End 4-Bit Inference on Generative Large Language Models " accepted at EMNLP24. (Github)
2024/04: I will join Apple as a Research Intern.
2024/02: I gave an invited talk on "The Art of LLM Compression" at Amazon.
2023/04: I will join Microsoft for a summer internship.
2023/01: "GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers" accepted at ICLR23.
2022/11: "ProbGraph: High-Performance and High-Accuracy Graph Mining with Probabilistic Set Representations " selected as the best paper at SC22.
2022/09: "ENS-10: A Dataset For Post-Processing Ensemble Weather Forecasts " accepted at Neurips22. (Github)