https://aryanputta.com/blog/ Notes on AI inference, distributed systems, and 6G networking. en-us https://aryanputta.com/blog/how-to-contribute-to-open-source-big-tech https://aryanputta.com/blog/how-to-contribute-to-open-source-big-tech Thu, 02 Jul 2026 09:00:00 GMT 17 merged PRs across NVIDIA, IBM, and HuggingFace, and not one came from a good first issue label. Why the labeled easy work is a race you lose, how to find bugs maintainers actually want fixed, and the PR habits that get a stranger's code shipped. https://aryanputta.com/blog/kv-cache-memory-wall kv-cache-memory-wall Tue, 23 Jun 2026 09:00:00 GMT The weights are not what run you out of GPU memory first. The KV cache is, and it scales with your traffic, not your model size. Here is how I think about that bottleneck and how EigenKache compresses it: 60.35% less memory, 19.87% less latency.