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.
Reducio-VAE is a model for encoding videos into an extremely small latent space. It is part of the Reducio-DiT, which is a highly efficient video generation method. Reducio-VAE encodes a 16-frame video clip to T/4∗H/32∗W/32…
TamGen is a transformer-based chemical language model for developing target-specific drug compounds. Research shows that TamGen can also optimize existing molecules by designing target-aware molecule fragments, potentially enabling the discovery of novel compounds that build…
TileIR (tile-ir) is a concise domain-specific IR designed to streamline the development of high-performance GPU/CPU kernels (e.g., GEMM, Dequant GEMM, FlashAttention, LinearAttention). By employing a Pythonic syntax with an underlying compiler infrastructure on top of…
Paper: “MMLU-CF: A Contamination-free Multi-task Language Understanding Benchmark”
RAD-DINO is a vision transformer model trained to encode chest X-rays using the self-supervised learning method DINOv2. RAD-DINO is described in detail in RAD-DINO: Exploring Scalable Medical Image Encoders Beyond Text Supervision (F. Pérez-García, H. Sharma, S.…
MAIRA-2 is a multimodal transformer designed for the generation of grounded or non-grounded radiology reports from chest X-rays. It is described in more detail in MAIRA-2: Grounded Radiology Report Generation (S. Bannur, K. Bouzid et al.,…
RadFact is a framework for the evaluation of model-generated radiology reports given a ground-truth report, with or without grounding. Leveraging the logical inference capabilities of large language models, RadFact is not a single number but a suite of…
This repository replicates the experiments of the paper “Cheap Permutation Testing”.
KBLaM is a new method for augmenting LLMs with external knowledge. Unlike Retrieval-Augmented Generation, KBLAM eliminates external retrieval modules, and unlike in-context learning, its computational overhead scales linearly with KB size rather than quadratically.
Magentic-One is a generalist multi-agent system created to address intricate web and file-based tasks. By utilizing an intelligent Orchestrator alongside specialized agents, it facilitates the automation of complex, multi-step activities across various environments.
