NFNets are faster than EfficientNets and they don’t use normalization — Our smaller models match the test accuracy of an EfficientNet-B7 on ImageNet while being up to 8.7× faster to train, and our largest models attain a new state-of-the-art top-1 accuracy of 86.5%. Source: arxiv One of the most annoying things about training a model is the time it takes to…

A comparison between the latest versions of PyTorch (1.8) and Tensorflow (2.5) — Tensorflow/Keras & Pytorch are by far the 2 most popular major machine learning libraries. Tensorflow is maintained and released by Google while Pytorch is maintained and released by Facebook. In this article, I want to compare them in terms of: What's new in the latest released versions Which one to…

A UCL project to solve the problem of tedious digital pathology labeling — In this article, I am going to talk about one of my dearest and best projects that I have worked on, my undergraduate thesis at University College London. My thesis was about using unsupervised machine learning to automate the process of breaking down huge Whole Slide Images and labeling them…

Evaluation, test coverage, and fine-tuning — In Part 1, we discussed open-source models, proprietary models, and model evaluation. This part is going to be more focused on further evaluation, “test coverage”, and fine-tuning. LLM Metrics It’s challenging to determine which metrics to measure for generative models, and there’s a focus on task-specific performance for traditional machine learning compared…

Open-source models, proprietary models, and model evaluation — Should you go with open-source or proprietary models? When it comes to language models, there is a lot of friction with open-source models due to licensing issues. While serving open-source models can create problems like difficulty in training and serving, proprietary models are easier to handle. However, there are cases where open-source models are needed for customization and…

A deep dive — data extraction, initializing the model, splitting the data, embeddings, vector databases, modeling, and inference — We are seeing a lot of use cases for langchain apps and large language models these days. After inspecting a lot of them and building a few myself, I wanted to write this article about the common concepts, ideas, and essentially the steps of building an LLM-langchain-powered application. Most of…

Perform question answering or summarisation from the comfort of Slack using Langchain on custom data. — Throughout my own journey of exploring LLMs (Large Language Models), I have made a tutorial for deploying your LLM app on Streamlit here. But then, a colleague of mine suggested simply adding it as a Slackbot. The UI is nicer, more chat friendly, and simply just easier to maintain, so…

A quick overview of hugging face transformer agents — Hugging Face has released a new tool called the Transformers Agent, which aims to revolutionize how over 100,000 HF models are managed. The tool provides a natural language API on top of transformers with a set of curated tools. The system supports both OpenAI modes and open-source alternatives. …

Fine-tuning is expensive and costly, do you really need it for simple question answering? — Many people ask how to train models on their Corpus of data to ask questions, and they assume fine-tuning is the way to go. However, fine-tuning is a type of transfer learning used to teach a model a new task, not new information. On the other hand, semantic search utilizes…