GRPO: How Language Models Learn to ReasonDo you remember the times when we used to make LLMs count the occurrence of a specific letter in a word, like "How many r's in strawberry?" Back then, LLMs used to get it wrong a lot of times, but now5 min read·swaritshukla.hashnode.devSep 22
Batch Size and LLM Inference EfficiencyHaving an optimal Batch size can decrease your models cost per token at the time of inference. This will be an explanation on how Batch size affects4 min read·swaritshukla.hashnode.dev21KJul 22
Nvidia 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 everyone7 min read·swaritshukla.hashnode.dev2May 4
The Elegance of MoE: How Gemma 4’s 26B Model Runs Like a 4B ModelHow can a 26B model run like a 4B one? Discover the power of Mixture of Experts (MoE) in Google's latest open-source model release5 min read·swaritshukla.hashnode.dev4Apr 12
