Nitin Gupta

Explains a technique to identify and share bit-identical weight tensors across models to reduce GPU memory usage for multi-model inference, reporting real-world VRAM savings and low overhead.

Explores the idea of loading a shared backbone of weights once on GPU and loading only unique pieces per model to save VRAM in self-hosted model deployments.

A practical guide covering eight commonly used JavaScript array methods (filter, map, find, forEach, some, every, reduce, includes) and how to use them effectively.

Technical write-up on improving higher-order allocation latencies in the Linux kernel via proactive background compaction; this work was later accepted into upstream kernel.