InTowards AIbySubodh Lonkar·15h agoLLM Inference & Serving at Scale: From First Token to 10,000 Concurrent UsersInside the engine room of modern AI: GPU memory, KV caches, PagedAttention, RadixAttention, vLLM, SGLang, latency, throughput, cost…
InTowards AIbySubodh Lonkar·2d agoBig Teacher, Small Student: Can a 0.6B Language Model Learn Financial Reasoning from an 8B Model?An end-to-end engineering journey through QLoRA, supervised fine-tuning, teacher-student distillation, hard-example mining, and targeted…
InTowards AIbySubodh Lonkar·4d agoI Built a 40M-Parameter Multilingual Language Model for Delivery Riders in 28 Minutes.From a free Colab GPU to a tiny Transformer: building a domain-focused Multilingual Small Language Model, understanding the fundamentals…A response icon1A response icon1
InThe StartupbySubodh Lonkar·Jan 28, 2021WSDM — KKBox’s Music Recommendation ChallengeA solution in the Top 0.7% for the Kaggle Competition
InAnalytics VidhyabySubodh Lonkar·Aug 27, 2020Dimensionality Reduction by Stochastic Gradient DescentThis post demonstrates the use of Stochastic Gradient Descent for Dimensionality Reduction.A response icon1A response icon1
InThe StartupbySubodh Lonkar·Aug 10, 2020Training an MLP From Scratch Using Backpropagation for Solving Mathematical EquationsThis post demonstrates the concept and use of backpropagation for solving a mathematical equation.
