My New Hugo Site

One cool feature in llama.cpp is the ability to provide a grammer from the LLM output, which has some powerful implications. This post is meant to show some examples of my work messing with python and llama.cpp grammars Format LLM response with a given JSON schema Lets say I want to generate API calls for a given prompt. I can define a JSON schema like this: schema = ''' { "$schema": "http://json-schema. [Read More]

I wanted to keep this as simple and cheap as possible. I write blog posts in a specific spot in my obsidian notes, and they get published to a netlify app. Implementation I found this cool project devidw/obsidian-to-hugo: Process Obsidian notes to publish them with Hugo which lets you create a static site from markdown. It worked great but I had to modify it to support a few requirements specific to my note system. [Read More]

What this website will be A place to share and organize my projects A place to share my ideas A home for my half baked ideas Half baked ideas One of my favorite college professors, Mattox Beckman, told a story one day of a computer scientist who decided to stop worrying about making everything a scientific paper worthy publication. Instead, he would start to share a plethora of mini papers and blog posts on his personal website, accumulating hundreds and hundreds of smaller posts. [Read More]

Recently Alfred version 5.5 was released which includes new UI elements and a cool chatgpt interface. By default this requires a paid chatgpt account to work, and you may want to use an alternative model for a variety of reasons such as: Cost Privacy Running models without gaurdrails Enhanced Customization Setup Ollama Ollama is an open source tool to download and run LLMs on your own machine. Download the installer from Ollama Run the installer in a terminal run: ollama run llama2 This will download the model file, which is about 4. [Read More]

Being obsessed with golf, and having unlimited access to several high end golf simulators, i’ve naturally spent a lot of time thinking how I can use my data-science and development experience to improve my golf game. Foresight golf simulators write all of their shot data to a nice JSON file, every time a shot is hit. This is great for data analytics, but for making some live training aids. Loading and Visualizing Shot Data To start, I made a simple streamlit dashboard that lets me see the dispersion pattern of each of my clubs. [Read More]