dylan lim
email
twitter
linkedin
hi! i'm a bs/ms student at stanford studying computer science. i work on ml systems. some of my work includes
compilers for inference optimization
, whole model kernel fusion (
low latency
,
high throughput
) and writing performant
multi-gpu communication primitives
.
prior work
we built megakernels, now we tore them down. tkcc's journey
(together ai 2026)
together compiler, continual compilation for AI inference
(together ai 2026)
to infinity and beyond, thunderkittens now on nvidia vera rubin
(together ai 2026)
we bought the whole gpu, so we're going to use the whole gpu
(hazy research 2025)
look ma, no bubbles! designing low-latency megakernels
(hazy research 2025)
one kernel for all your gpus
(hazy research 2025)
layoutvlm: 3d scene generation
(cvpr 2025)
flexflow: distributed dnn compiler
(stanford 2024-25)
things i built
shard — distributed training system for local compute
flexgraph — ml model architecture generation for compiler evaluation
industry
together ai — polymorphizing
(present)
jump trading — core strategies
(summer 2025)