10 minutes to 33 seconds
EvryStay's backend tests, and the things we tried that didn't work.
Writing
Notes on AI infrastructure, product engineering, and whatever I’m building this week.
Optimisations Wednesday, 25% more throughput than Ollama
Smarter & deeper MTP, MTPLX, and less CPU/GPU synchronisation on Qwen.
Sixteen chips just lost to eight
My paper "Less is More" has been accepted for presentation at the BID Workshop at ICPP 2026.
DeepNeuron honoured in Japan for APAC HPC-AI success
Monash DeepNeuron placed equal third in the 8th APAC HPC-AI Competition, competing against 49 university teams.
A field guide to Aspire 2A+
CUDA, modules, PBS Pro, storage, and build notes for NSCC Singapore's H100 cluster.
A field guide to Massive 3
Scratch storage, Slurm, caches, and module notes for Monash University's M3 cluster.
SGLang and DeepSeek
Installing, serving, and benchmarking DeepSeek-R1 with SGLang across competition clusters.
SGLang optimisations that moved throughput
The flags, couplings, failed NCCL experiments, and parallel layouts behind the final result.