Dataset Source Code
BAPO – Bounded Attention Prefix Oracle
This repository contains all scripts for re-producing the results of our paper “Lost in Transmission: When and Why LLMs Fail to Reason Globally”.
Researcher 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.
Tool
ExACT
ExACT is an approach for teaching AI agents to explore more effectively, enabling them to intelligently navigate their environments, gather valuable information, evaluate options, and identify optimal decision-making and planning strategies.
Tool
Microsoft Research Accurate Chemistry Collection (MSR-ACC)
The Skala functional will enable more accurate, scalable predictions in computational chemistry. It starts with the largest high-accuracy dataset ever built for training deep-learning-based density functional theory (DFT) models. This dataset underpins Skala—coming soon to…
Dataset Source Code
Science Foundation Model
We develop the Science Foundation Model to empower natural scientists and accelerate breakthroughs in scientific discovery. As part of this effort, we introduce the sequence-based model, Nature Language Model (NatureLM). NatureLM is designed to span…
Dataset Source Code
EfficientXLang
This codebase is the official implementation of “EfficientXLang: Towards Improving Token Efficiency Through Cross-Lingual Reasoning.”
Tool
Phi-4
Phi-4-multimodal and Phi-4-mini, the newest models in Microsoft’s Phi family of small language models (SLMs) are now available. These models are designed to empower developers with advanced AI capabilities. Phi-4-multimodal, with its ability to process…
Tool
ReMe
ReMe is a web-based framework that helps researchers create AI chatbots for personalized training and interventions aimed at strengthening memory and cognitive functions. Early evaluations show its potential to contribute to digital health innovation and…
Tool
EvoDiff
EvoDiff is a general-purpose diffusion framework that combines evolutionary-scale data with the distinct conditioning capabilities of diffusion models for controllable protein generation in sequence space. EvoDiff generates high-fidelity, diverse, and structurally-plausible proteins that cover natural…
Tool
PEACE
PEACE enhances multimodal large language models (MLLMs) with geologic expertise, enabling accurate interpretation of complex, high-resolution maps. By integrating structured extraction, domain knowledge, and reasoning, it supports critical tasks in disaster risk, resource discovery, and…