Deep learning and hacking

Deep learning and hacking

Self-hosting Mattermost on AWS Lightsail

If you haven't heard of Mattermost, it's an open source communications platform similar to Slack. It has a gorgeous UI: and it also comes with native mobile apps that can connect to self-hosted versions. Unlike the free version of Slack, there are no limitations on how long messages will be

Setting up Let's Encrypt SSL on Nginx

Let's say you have an nginx site that's proxying to a node.js backend, so your config looks like this: server { listen 80; server_name app1.example.com location / { auth_basic "Restricted Access"; # Authentication realm auth_basic_user_file /etc/nginx/.htpasswd; # Path to the password proxy_pass http://127.

Downloading Llama2 from hugging face

For the current project I'm working on, I'm trying to fine-tune the retriever in a RAG pipeline using the Arcee/DALM repo. This requires passing in Meta's Llama2-7B model as the generator that is used in the fine-tuning process. Here are the steps in case it helps anyone! Step 1:

Detecting and shutting down idle Sagemaker instances

As an indie hacker on a low budget, I want to minimize my cloud spend when doing deep learning experimentation. Since GPUs are quite costly, it's crucial to avoid having them sit idle. It's very easy to kick off a training job that you think will run all night, have

Building a PDF chat tool: A Technical Deep Dive

Recently, our team at Lazy Consulting embarked on an exciting project: constructing a web tool that enables interaction with PDF documents via LLMs. First, a quick demo! Requirements We set out to build something that would meet the following requirements: 1. Enable uploading of custom documents (PDF format only) 2.

Tennis court line detector: Part 5, test on unseen tennis court

Tennis court line detector: Part 5, test on unseen tennis court

In part 4 we got fairly good results on the validation set. In this part, we'll try to run the validation on this new synthetically generated tennis court, which is not present in the training data. Here's a 3rd tennis court model, this time with a grass surface. For the

Tennis court line detector: Part 4, generate validation and test data

Tennis court line detector: Part 4, generate validation and test data

Add more tennis courts I created a brand new Unity project, added this tennis court, and went through the same steps as described in Part 1 to add the keypoints. This was tedious, I wish I could find a better way! To avoid confusing the network, I "grayed out" the

Tennis court line detector: Part 3, detecting non-visible keypoints

Tennis court line detector: Part 3, detecting non-visible keypoints

In part 2 we got a basic CNN trained to overfit the training data, but all of the training data showed all keypoints as being visible. Generate images with non-visible keypoints Let's make things a bit harder for the network and generate images where some of the keypoints are off

Tennis court line detector: Part 2, training a CNN

Tennis court line detector: Part 2, training a CNN

Synthetic data settings To recap part 1, we built a synthetic data generator for tennis courts with labeled keypoints. with corresponding annotations in the SOLO JSON format. Here are the randomizer settings: Note that the rotation randomizer only makes relatively small rotations to spin the court on the y axis

Tennis court line detector: Part 1, synthetic data generation

Tennis court line detector: Part 1, synthetic data generation

Disclaimer: unfortunately, this project did not produce decent results. I decided to publish it anyway, because there are a few nuggets of knowledge I learned along the way that I wanted to share to other computer vision enthusiasts. I'm also holding onto some hope that someone will come along and

A VSCode epiphany

A VSCode epiphany

I ran into an issue on Jetbrains CLion that prevented the debugger from working, so I decided to give VS Code for a spin for C++ development. This post will cover a few topics: 1. Running and debugging CMake projects in VS Code 2. Recognizing include directories for 3rd party

AWSim + autoware running on Ubuntu 20.04

AWSim + autoware running on Ubuntu 20.04

My first attempt to run AWSim (autoware simulator) from inside a docker container didn't go so well: $ ./AWSIM.x86_64 --help Found path: /../AWSIM/AWSIM.x86_64 Segmentation fault (core dumped) Here's a walkthrough of how I worked around it. Retry on host I initially attempted the command inside of

Install autoware from source

Install autoware from source

Previously I walked through installing autoware in a docker container, but I've run into an issue trying to run AWSim within a container, and so I decided to install autoware from source rather than running it inside a docker container. This post is mainly about the stumbling blocks I hit

Installing autoware + docker (cont'd)

Installing autoware + docker (cont'd)

This is a continuation of Installing Autoware on Ubuntu 20.04. I'm starting over after working with the system76 support engineers to address some issues I ran into. The tl;dr is that I ended up having a mix of packages from the system76 apt repositories and the nvidia apt

Career pivot: applied ML 🤖

Career pivot: applied ML 🤖

I've recently decided to pivot my career into applied ML, more specifically applied Deep Learning. I'm planning to document my journey, so if you're taking the same journey or interested in doing so, check my twitter feed for regular updates. Rather than trying to juggle this with another job, I'm