
MatterGen is a generative model from Microsoft for designing inorganic materials across the periodic table. It can be fine-tuned to steer generation toward property constraints.
Web appFree
Step 1: Install with uv on Linux with a CUDA GPU, and set up Git LFS.
Step 2: Pick a checkpoint such as the base model or a fine-tuned one.
Step 3: Run mattergen-generate with a results path, model name and batch size.
Step 4: Add properties_to_condition_on and a diffusion guidance factor for conditional generation.
Step 5: Run mattergen-evaluate on the results to get metrics in a metrics.json file.
The install steps assume Linux and a CUDA GPU. Running on Apple Silicon is marked experimental.
It controls how closely generated samples follow the input property values. Zero is unconditional generation, and higher values trade diversity and realism for adherence.
No. The README says they were re-trained with this repository, so results may differ slightly from the publication.
The evaluation command relaxes them with the MatterSim force field and computes novelty, uniqueness and stability. You can instead supply energies from another method such as DFT.
MatterSimMatterSim is a deep learning atomistic model that works across elements, temperatures and pressures. It is distributed as a Python package with pre-trained models.
BiomedCLIPBiomedCLIP is a biomedical vision-language model from Microsoft, pretrained on 15 million figure-caption pairs from PubMed Central. It supports retrieval, image classification and visual question answering research.
AuroraAurora uses AI for AI foundation model for Earth-system prediction across weather, air quality, ocean, and environmental forecasting tasks.
ChempropChemprop is a PyTorch-based framework for training and evaluating message-passing neural networks that predict molecular properties. The site hosts its documentation for version 2.3.1.
DeepChemDeepChem is an open-source Python library for machine learning and deep learning on molecular and quantum datasets. It builds on PyTorch and other popular frameworks.