followfox.ai’s Newsletter

Followfox.ai's Newsletter is a comprehensive exploration into small AI models, focusing on their development and optimization for edge computing. It discusses various AI technologies such as Stable Diffusion, LLaMa, and Vodka models, offering guidance on fine-tuning, learning rates, and running AI on limited resources. The blog advocates for open-source development, reflects on AI's potential impacts on industries, and experiments with machine learning techniques.

AI Model Development and Optimization Open-Source AI Technologies Machine Learning Techniques AI in Edge Computing Industry Impacts of AI

The hottest Substack posts of followfox.ai’s Newsletter

And their main takeaways

Getting Started with LoRAs + Vodka Photorealism

176 implied HN points • 15 Jun 23
🕹 Technology AI Training Modeling
  1. The post discusses getting started with LoRAs and creating a photorealistic LoRA for Vodka models.
  2. It includes steps like downloading and using a LoRA, training the first LoRA, and finally fine-tuning a custom LoRA for photorealistic results.
  3. The process involves using specific tools, datasets, and parameters to train LoRAs, and explores possibilities for creating high-quality, realistic images.

Find Optimal Learning Rates for Stable Diffusion Fine-tunes

157 implied HN points • 13 Mar 23
🕹 Technology Machine Learning Data Analysis Experimentation Optimization Model Training
  1. Estimate the minimum and maximum learning rate values by observing when the loss decreases and increases during training.
  2. Choosing learning rates within the estimated range can optimize model training.
  3. Validating learning rate ranges and fine-tuning with different datasets can improve model flexibility and accuracy.

Getting Started with ComfyUI on WSL2

157 implied HN points • 10 Apr 23
🕹 Technology AI Installation Tools Generative AI
  1. Consider exploring ComfyUI as an alternative to Automatic1111 for Stable Diffusion.
  2. Installing ComfyUI on WSL2 involves setting up WSL2, installing CUDA, Conda, and git, cloning the repo, and running tests.
  3. After installation, experiment with different modules, compare outputs with Automatic1111, explore examples in the repo, and share findings.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

How to run LLaMa in an old GPU

137 implied HN points • 20 Apr 23
🕹 Technology AI Software Hardware Development Installation
  1. Local LLMs are not as advanced as ChatGPT but show potential for various applications.
  2. LLaMa models by Facebook are licensed for non-commercial use and show good performance for their size.
  3. GPTQ quantization technique enables running LLaMa on old GPUs by compressing model weights and maintaining speed.

Partying like it's 1999

137 implied HN points • 16 Mar 23
🕹 Technology AI Music Industry
  1. Innovation often comes from unexpected intersections of technologies and fields, like how music sharing led to cryptocurrencies.
  2. Technological progress is a complex web of interactions with surprises and unintended consequences.
  3. The future of AI holds unpredictable surprises and revolutionary advancements, much like the evolution from music sharing to cryptocurrencies.

Open Source and Open Seas — or: Why AI will be Open Source

117 implied HN points • 03 Jun 23
🕹 Technology Open Source AI Software Development Generative Art Tech industry
  1. Open source software has become a foundational layer of innovation and is prevalent in tech stacks globally.
  2. The interest in open source stems from its ease of debugging, fostering innovation, and being cost-free.
  3. The evolution of tech industries, like AI, shows a progression towards open source to drive innovation and accessibility.

D-Adaptation: Goodbye Learning Rate Headaches?

98 implied HN points • 21 Jun 23
🕹 Technology Artificial Intelligence Machine Learning Optimization Comparative Analysis
  1. D-Adaptation method automates setting learning rate, aiming for optimal convergence in machine learning.
  2. Implementing D-Adaptation can consume more VRAM and result in slower training speed compared to other optimizers.
  3. Initial results show D-Adaptation performing comparably to hand-picked parameters in generating high-quality models.

Releasing Vodka V2 and All the Details How We Made it [Part 2]

117 implied HN points • 18 May 23
🕹 Technology AI Model Training Image Generation Machine Learning
  1. Vodka V2 was released with an updated dataset and marginally better model compared to V1
  2. The key changes in V2 included using a better dataset, increasing data volume, and cleaning the data more thoroughly
  3. The training protocol for V2 involved lower learning rate and enhanced data cleaning to achieve smoother training and optimize model performance

[May 2023] Latest Automatic1111 Installation with WSL2 on Windows

117 implied HN points • 05 May 23
🕹 Technology Installation Updates Guides Dependencies
  1. The latest version of AUTOMATIC1111 includes torch 2.0.0 and ControlNet 1.1, showing the repo is active and relevant
  2. You can choose between using a prebuilt FollowFox.AI WSL distribution for an easy setup or manually installing on WSL2 with steps like setting up CUDA and Conda
  3. Installing ControlNet 1.1 is essential for new features and options, requiring downloading and installing the .pth files for use in the updated Auto1111

Saving Time Using Auto1111's API: Automated Workflow for XYZ Grids

98 implied HN points • 27 May 23
🕹 Technology Automation API Workflow Testing
  1. An automated workflow using Auto1111's API can save time when generating XYZ comparison grids
  2. The process involves creating a CSV file with parameters for each grid and using a script to feed these parameters to Auto1111 through the API
  3. While this automated workflow can save time, it may not allow for immediate review and adjustments after each grid generation