Notes on tech, code, and building thingshttps://blog.vaishak.me/https://blog.vaishak.me/posts/gemma4-dgx-spark/https://blog.vaishak.me/posts/gemma4-dgx-spark/Every inference stack on GB10 hardware — llama.cpp, Ollama, vLLM — across quantizations and concurrencies. Model architecture accounts for ~8x of the throughput difference.Sun, 12 Apr 2026 00:00:00 GMThttps://blog.vaishak.me/posts/hello-world/https://blog.vaishak.me/posts/hello-world/A quick intro — who I am, what I work on, and what this blog is about.Sun, 12 Apr 2026 00:00:00 GMT