Yesterday





These days, I’m exploring the field of natural language generation, using auto-regressive models such as GPT-2. HuggingFace transformers offers a host of pretrained language models, many of which can be used off the shelf with minimal fine-tuning. In this post, however, we will try to build a small GPT model from scratch using PyTorch. I’ve realized that sometimes I feel eerily insecure about using things without truly understanding how they work. This post is an attempt at understanding the GPT model. Turns out that its implementation is actually quite similar to the transformers implementation I wrote about in an earlier post, so you will see some repeats here and there. Also helpful was Karpathy’s minGPT repository, which I heavily referenced for this particular implementation. With that sorted, let’s get started!

by rmuell9 1 day ago saved 2 times



A cozy command-line clubhouse for computer people. Chat, music, games, art, coding, and tech news. Connect with any SSH client!

by isaac 1 day ago saved 5 times


These are some products that I’m excitedly following (or am keen to get) because they’re technologies without the intent to dominate.


Free Ink is an open-source collective building the software, firmware and hardware for e-paper readers — every layer ships in the open for anyone to pick up, extend and make their own.

by srivathsan 1 day ago saved 2 times



Continuing his quest for a healthier, more handsome Hitch, the author puts himself in the hands of four experts. Yes, a Brazilian wax was involved.


There’s an entire micro-economy based on the pursuit of betterment. The author—58, full-figured, and ferocious in his consumption of cigarettes and scotch—agreed to test its limits, starting with the Executive De-Stress Treatment at a high-end spa.