Last week I was running some computer vision inference benchmarks when I heard my machine’s fans instantly fire up. It sounded like a rocket ship taking off, and it prompted me to take a peek at htop and view some details on my machine’s processes. Huh, that’s strange. Out of my 20 CPU cores, 14 of them were pegged at ~100% utilization, and when I killed my benchmark, they all shot back to ~0%....
Recently I’ve been interested in some of the ways that we can use embeddings to compare images, text, and other data. Embeddings allow us to create features and applications like image search, retrieval-augmented generation, and many more. You can think of an embedding as just a list of numbers (a vector) that represents the semantic meaning of a given piece of data. this allows us to compare two embeddings and determine how similar they are....
I’ve been using Obsidian to take notes now for a little over a year. I’ve fallen in love with the tool and it has really helped me organize my thoughts, work meetings, blogs, and general todo lists. I’ve noticed one problem that has become somewhat annoying recently though. I’ve found that I have a lot of meeting notes that I’ve started, but I haven’t formatted and organized them well. This is no fault of the tool by the way....
As I continue to work through chapters of the Practical Deep Learning course, my game plan is to write blog posts to solidify my learning of the content. Next on the list is taking our drink classifier we built in the first chapter and deploy it to production. There are many ways to go about doing this, but in this blog we’ll deploy to Hugging Face with Gradio. Let’s do this 🥂....
This blog is a short one but has improved my search experience immensely. Last week I heard a lot of interesting content around a feature from Arc called Arc Search. Arc Search Apparently it’s a wrapper around Perplexity AI that enables a more interactive and creative search experience. If you haven’t seen it in action, here’s a review of the app on YouTube. After doing some googling though, I was disappointed to find that there wasn’t a desktop mac version of Arc Search....
I recently started a new course called Practical Deep Learning. The course is taught by Jeremy Howard, a machine learning legend and founder of fast.ai. One of the first lessons in the course was training a classifier to detect birds in images. As part of learning the materials, I thought i’d be a fun experiment to extend this model to classify the top 10 most popular cocktails. Let’s get started 🍸🍹....
It’s been 2 days since the launch of Llama 3 and I’ve been itching to try it out. Here is a short blog post experimenting with Ollama to locally run this sota (state of the art) LLM (large language model). But first, what is Llama 3? What is Llama 3 Meta’s Llama 3 is the most capable openly available LLM as of writing this blog. On April 18th, 2024, Meta released two models of this generation with 8B and 70B parameters....
Today I was playing around with an LLM called Mistral 7B by running it locally with Ollama. Once installed, Ollama provides a chat interface and an API that you can use and run wherever. curl http://localhost:11434/api/generate -d '{ "model": "mistral", "prompt":"tell me a joke?" }' When running this API call, I noticed that responses were streamed back to the client in a way that appears to be token by token. Take a look at running the command....
Today’s post is about creating terminal GIFs. GIFs (Graphics Interchange Format) were invented in 1987 by an American computer scientist named Steve Wilhite. Although the pronunciation has been the rife with memes and hot debate, I thought it’d be a fun to specifically find out an easy way to create GIFs of my terminal for blog posts. I was writing a post about streaming LLM requests in Python and a GIF would provide such a clean display of the idea I was trying to get across....