SSSwarit Shuklainswaritshukla.hashnode.dev·May 4 · 7 min readNvidia Vs Everyone: Why everyone is building their own acceleratorsThere has been a race in the AI industry where companies are now preferring their own custom accelerators to train AI models and generate inference from them. But why? Why all of a sudden is everyone 10
SSSwarit Shuklainswaritshukla.hashnode.dev·Apr 12 · 5 min readThe Elegance of MoE: How Gemma 4’s 26B Model Runs Like a 4B ModelGoogle recently dropped its new family of open-source AI models, Gemma 4, but the variant that truly captured my interest is Gemma-4-26B-A4B-IT. The question is: how can a 26 billion parameter model o30
