P2P Docker registry capable of distributing TBs of data in seconds

The Art of Debugging

Machine Learning Engineering Open Book

Machine Learning From Scratch. Bare bones NumPy implementations of machine learning models and algorithms with a focus on accessibility. Aims to cover everything from linear regression to deep lear…

A simple web app for NUS students to compute their GPA and keep track of courses taken on an Excel sheet.

Designing a Dashboard for Transparency and Control of Conversational AI, https://arxiv.org/abs/2406.07882

(NeurIPS 2024 Oral 🔥) Improved Distribution Matching Distillation for Fast Image Synthesis

Benchmarking long-form factuality in large language models. Original code for our paper "Long-form factuality in large language models".

This repository contains the dataset and code for "WiCE: Real-World Entailment for Claims in Wikipedia" in EMNLP 2023.

A package to evaluate factuality of long-form generation. Original implementation of our EMNLP 2023 paper "FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation"

Tools for understanding how transformer predictions are built layer-by-layer

Hello World

Sparsify transformers with SAEs and transcoders

Code for our paper "Attending to Graph Transformers"

Create feature-centric and prompt-centric visualizations for sparse autoencoders (like those from Anthropic's published research).

Training Sparse Autoencoders on Language Models

Sparse Autoencoder for Mechanistic Interpretability

Codebase for reproducing the experiments of the semantic uncertainty paper (short-phrase and sentence-length experiments).

Code and Data Repo for the CoNLL Paper -- Future Lens: Anticipating Subsequent Tokens from a Single Hidden State

LLM training in simple, raw C/CUDA

DimeNet and DimeNet++ models, as proposed in "Directional Message Passing for Molecular Graphs" (ICLR 2020) and "Fast and Uncertainty-Aware Directional Message Passing for Non-Equilibrium Molecules…

MACE - Fast and accurate machine learning interatomic potentials with higher order equivariant message passing.

Official implementation of Graph Neural Networks as Gradient Flows.

Open-Sora: Democratizing Efficient Video Production for All