This post walks through a KV Cache misalignment issue for encoder–decoder models (Whisper, Llama-3.2-Vision/Mllama) on Intel XPU in SGLang. We will start our discussion with how SGLang manages its KV cache and then discuss the issue with possible fixes.

A practical deep dive into how PyTorch tensors use storage, stride, views and contiguity.

Explaination of different Paramete Efficient Finetuning Techniques

A brief summary of PyTorch's implementation of Distributed Data Parallel DDP.

A brief note on ZeRO's workings.

A brief summary of different distributed training strategies used to train LLMs.

In this article, I will give you a brief overview of Named Entity Recognition (NER), its importance in information extraction, few approaches to perform NER, and at the end will also show you how to implement NER as an MRC problem.

Let us dive into the BERT's architecture and details of formulating Question Answering NLP task for transformer models.

In this article I will discuss an efficient abstractive text summarization approach using GPT-2 on PyTorch with the CNN/Daily Mail dataset.

YOLO has been a very popular and fast object detection algorithm, but unfortunately not the best-performing. In this article I will highlight simple training heuristics and small architectural changes that can make YOLOv3 perform better than models like Faster R-CNN and Mask R-CNN.