🏠

Blogs

  1. Finetune LLM with QLoRA

    Brief intro to QLoRA with concepts and practical implmentation of finetuning a LLM using QLoRA

  2. Image Segmentation and Inpaint

    Leveraging U-Net for human segmentation to generate a precise mask to isolate humans from an image and use inpainting to edit image backgrounds

  3. Basics of LoRA and its implementation

    Brief intro to LoRA and its working with practical implementation explained

  4. Explore Data Augmentation with OpenCV

    This blog explores using OpenCV for custom data augmentation to enhance model training for photo restoration tasks. It highlights how synthetic variations like noise and aging effects can improve model robustness despite limited labeled data.

  5. Fully Sharded Data Parallel (FSDP)

    This blog provides a clear and concise explanation of Fully Sharded Data Parallel (FSDP), a method for memory-efficient and scalable training of large models. It covers the fundamentals of how FSDP works, compares the original implementation (FSDP1) with the improved FSDP2, and highlights the key benefits of using FSDP in modern deep learning workflows.