Posted inArtificial Intelligence Python PyTorch Web Dev Offload LLM using Facebook FairScale Posted by By Jacob K. Lo October 1, 2023 What is Fairscale FairScale is an open-source library from Facebook. It help researchers and engineers…
Posted inArtificial Intelligence Python PyTorch DeepSpeed API ZeRO Offload Posted by By Jacob K. Lo September 12, 2023 The biggest problem in Deep learning and training large language models (LLMs) is the limit…
Posted inArtificial Intelligence Python PyTorch Offload LLM custom back propagation Posted by By Jacob K. Lo September 2, 2023 What is Model Offloading? Model offloading, as described in the DeepSpeed ZeRO's paper, refers to…
Posted inArtificial Intelligence Wandb get started Posted by By Jacob K. Lo August 23, 2023 Weights & Biases (W&B) is an MLOps platform, equipped with tools to help you visualize…
Posted inArtificial Intelligence PyTorch How to generate prompts with 4x less GPU memory Posted by By Jacob K. Lo August 19, 2023 What is 32-bit full precision training and 16-bit half precision training? This picture pretty much…
Posted inArtificial Intelligence Python PyTorch NanoGPT – My takes Posted by By Jacob K. Lo August 5, 2023 NanoGPT is develop by Andrew Karpathy, a Lead engineer from Tesla. The simplest, fastest repository…
Posted inPython Web Dev Visual Studio Code debug settings Posted by By Jacob K. Lo July 21, 2023 Include environment { // Use IntelliSense to learn about possible attributes. // Hover to view…
Posted inArtificial Intelligence Matplotlib Top 5 features on MatPlotlib for Deep Learning Posted by By Jacob K. Lo July 7, 2023 Matplotlib is an incredibly powerful tool for visualization that no data engineer should overlook. Before…
Posted inPython Useful Anaconda commands Posted by By Jacob K. Lo June 28, 2023 Create environment If you're juggling multiple projects like I am, keeping your packages and dependencies…
Posted inArtificial Intelligence Python Tensorboard summary writer 101 Posted by By Jacob K. Lo June 22, 2023 Tensorboard is a very convenient way to record a neural network's weights and parameters while…