Sriman Achanta
Machine Learning Systems and Performance ยท Virginia Commonwealth University
About
I am an undergraduate studying computer science at Virginia Commonwealth University. My research is in machine learning systems and performance, with a focus on making large language model training and inference faster and more predictable on modern GPUs.
Most recently I developed FoldAttention, an attention kernel for fast long-context decoding and deterministic training. I also build cuNIBS, a GPU finite element solver for transcranial magnetic stimulation, in VCU's Biomagnetics Laboratory under Dr. Ravi L. Hadimani.
Research Interests
- Machine learning systems and performance: GPU kernels for LLM training and inference
- Efficient LLM inference: attention, KV-cache decoding, and long-context serving
- Deterministic and reproducible training at scale