Dataset Source Code
LoRA
This repo contains the source code of the Python package loralib and several examples of how to integrate it with PyTorch models, such as those in HuggingFace. We only support PyTorch for now. See our…
Research Tools: code, datasets, & models
Discover an index of datasets, SDKs, APIs and open-source tools developed by Microsoft researchers and shared with the global academic community below. These experimental technologies—available through Azure AI Foundry Labs (opens in new tab)—offer a glimpse into the future of AI innovation.
Dataset Source Code
MoLeR: A Model for Molecule Generation
Implementation of MoLeR: a generative model of molecular graphs which supports scaffold-constrained generation. This open-source code accompanies our paper “Learning to Extend Molecular Scaffolds with Structural Motifs”, which has been accepted at the ICLR 2022…
Tool
Project Iris
Github link to Iris – pretrained summarization models for structured datasets and cardinality estimation.
Dataset Source Code
AMOS
Pretraining Text Encoders with Adversarial Mixture of Training Signal Generators. This repository contains the scripts for fine-tuning AMOS pretrained models on GLUE and SQuAD 2.0 benchmarks. Accompanying paper accepted to ICLR 2022: “Pretraining Text Encoders…
Dataset Source Code
Ekya: Continuous Learning on the Edge
Ekya is a system which enables continuous learning on resource constrained devices. Given a set of video streams and pre-trained models, Ekya can continuously fine-tune the models to maximize accuracy by intelligently allocating resources between…
Dataset Source Code
KID: Knowledge Infused Decoding
Knowledge Infused Decoding (KID) is a decoding algorithm that infuses knowledge (from Wikipedia) into each step decoding of text generation.
Dataset Source Code
Jigsaw Datasets
Jigsaw Dataset: Natural language to Python Pandas code. Two datasets (PandasEval1 and PandasEval2) described in our paper, “Jigsaw: Large Language Models meet Program Synthesis”.
Dataset Source Code
MSCCL: Microsoft Collective Communication Library
Microsoft Collective Communication Library (MSCCL) is a platform to execute custom collective communication algorithms for multiple accelerators supported by Microsoft Azure.
Dataset Source Code
Blue whale acoustics
The goal of this project is to use audio recordings and corresponding annotations to build an automatic classifier for calls from four different species of blue whales, and to estimate the total number of calls…
Dataset Source Code
Maximal Update Parametrization (ÎĽP)
Maximal Update Parametrization (ÎĽP) and Hyperparameter Transfer (ÎĽTransfer), in association with the paper: Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer