Daniel Han

cuML - RAPIDS Machine Learning Library

sciblox - Easier Data Science and Machine Learning

Material for gpu-mode lectures

Tool for generating high quality Synthetic datasets

Sometimes we want to ”reverse” a Markov Chain process. Taking the inverse of the transition matrix allows this to work, but the inverse result is not a transition matrix. If I wanted to model a population going to work, and then going back home, negative and greater than 1 probabilities in the inverse matrix will cause issues. I propose a method to compute the ”inverse” of a transition matrix, and the result is still a transition matrix

A high-throughput and memory-efficient inference and serving engine for LLMs

Inference, Fine Tuning and many more recipes with Gemma family of models

Data Science Society of UNSW Lecture Series

We developed a data mining solution to scrape data from multiple sources to get the latest and greatest information about health grants. AND - we designed a cool Jupyter Notebook Solution to visualise our 2nd challenge- to get data constantly from a source, and show it in Plotly

Created a Machine Learning system to predict if someone is going to suicide, and proposed some mechanisms to counter and help potential victims.

SkyPilot: Run AI and batch jobs on any infra (Kubernetes or 12+ clouds). Get unified execution, cost savings, and high GPU availability via a simple interface.

Clean, accessible reproduction of DeepSeek R1-Zero

LLM inference in C/C++

We used NSW Air Quality data, health data, transport OPAL congestion data and also some novel disease spreading algorithms to try model and find correlations between Air Quality, Congestion and Influenza rates.

No description.

A collection of machine learning notes

Accessible large language models via k-bit quantization for PyTorch.

We created a new Health Rating system that aggregated data from QANTAS, ABS, Victoria Health and supermarket data to reveal correlations between them. We got Special Commendations for our great work!

No description.

No description.

efficient arrays of booleans for Python

🤗 PEFT: State-of-the-art Parameter-Efficient Fine-Tuning.

The official PyTorch implementation of Google's Gemma models

RAPIDS Sample Notebooks

I noticed that traditional methods to predict a disease outbreak was by performing sentiment analysis on Twitter posts and Google Search terms. Unfortunately, these methods were inadequate, as Twitter and Google is not popular in all countries. So, I created a system to model and predict outbreaks without the need for social media. The system was able to update the probabilities of a virus from spreading from A to B in real time, and I plan to release it to the public next year. I also used Machine Learning and Deep Learning to predict larger long-term virus trends with Google Trends, and this acted as a validator for the MSIRD model.

Finetune Llama 3.1, Mistral, Phi & Gemma LLMs 2-5x faster with 80% less memory

Utils for Unsloth https://github.com/unslothai/unsloth

No description.

Sphinx theme for readthedocs.org

Fast, flexible and easy to use probabilistic modelling in Python.

NumPy-like API accelerated with CUDA

GPU Accelerated t-SNE for CUDA with Python bindings

No description.

We're creating a full Earth simulation to predict the future of everything and make JARVIS a reality!

Project Aeros aims to optimise data science processes in Python, and provides the user with easy to use functions and APIs

Finetuning LLaMA with cleaned datasets

No description.

No description.

Llama 2 Everywhere (L2E)

🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.

An interface library for RL post training with environments.

BLAS-like Library Instantiation Software Framework

No description.

Statsmodels: statistical modeling and econometrics in Python

Convert HTML to Markdown

A PyTorch Extension: Tools for easy mixed precision and distributed training in Pytorch

Train transformer language models with reinforcement learning.

No description.

Efficient Triton Kernels for LLM Training

No description.

Public repo for HF blog posts

Throwaway staging: validate CUDA 13.3 prebuilt build leg