Hugging Face's logo
ibm-research 's Collections
REAL-MM-RAG-Bench_BEIR
AI-Agent-4-Industry-4.0
Otter-Knowledge
REAL-MM-RAG-Bench
Granite 3.2 Models (GGUF)
Materials

Materials

Welcome to IBM’s multi-modal foundation model for materials, FM4M, designed to support and advance research in materials science and chemistry.

Upvote
5

    Note A Multi-view Mixture-of-Experts framework designed to predict molecular properties by integrating latent spaces derived from SMILES, SELFIES, and molecular graphs. Our approach leverages the complementary strengths of these representations to enhance predictive accuracy.

    Note 3DGrid-VQGAN is an encoder-decoder chemical foundation model for representing 3D electron density grids, pre-trained on a dataset of approximately 855K molecules from PubChem database. 3DGrid-VQGAN efficiently encodes high-dimensional data into compact latent representations, enabling downstream tasks such as molecular property prediction with enhanced accuracy.

    Note A Position-based Equivariant Graph Neural Network foundation model for Chemistry and Materials. The model was pre-trained on 1.4M samples (i.e., 90%) from the Materials Project Trajectory (MPtrj) dataset to predict energies, forces and stress. pos-egnn can be used as a machine-learning potential, as a feature extractor, or can be fine-tuned for specific downstream tasks.

    Note We present MHG-GED, an autoencoder architecture that has an encoder based on GNN and a decoder based on a sequential model with MHG. Since the encoder is a GNN variant, MHG-GNN can accept any molecule as input, and demonstrate high predictive performance on molecular graph data. In addition, the decoder inherits the theoretical guarantee of MHG on always generating a structurally valid molecule as output.

    Note SELFIES-TED introduces a transformer trained on SELFIES strings for improved molecule property prediction. SELFIES-TED uses a BART backbone to learn a molecule representation while also being able to generate novel molecules. SELFIES-TED has 2.2M parameters and was trained on >1 billion molecules from zinc-22, applying smiles enumeration.

    Note SELFIES-TED introduces a transformer trained on SELFIES strings for improved molecule property prediction. SELFIES-TED uses a BART backbone to learn a molecule representation while also being able to generate novel molecules. SELFIES-TED has 354M parameters and was trained on 1 billion molecules from zinc-22, applying smiles enumeration.

    Note A Mamba-based encoder-decoder chemical foundation model, SMILES-based State-Space Encoder-Decoder (SMI-SSED), pre-trained on a curated dataset of 91 million SMILES samples sourced from PubChem, equivalent to 4 billion molecular tokens.

    Note Note A large encoder-decoder chemical foundation model, SMILES-based Transformer Encoder-Decoder (SMI-TED), pre-trained on a curated dataset of 91 million SMILES samples sourced from PubChem, equivalent to 4 billion molecular tokens. SMI-TED supports various complex tasks, including quantum property prediction. Our experiments across multiple benchmark datasets demonstrate state-of-the-art performance for various tasks.

    Note Explore Foundation Models for Materials with an intuitive Gradio app! Test our state-of-the-art models—SMI-TED, SELFIES-TED, and MHG-GED—on your custom datasets for both classification and regression property prediction tasks. Get insights into materials science with ease

    Note Explore Foundation Models for Materials with an intuitive Gradio app! Test our state-of-the-art models—SMI-TED, SELFIES-TED, and MHG-GED—on your custom datasets for both classification and regression property prediction tasks. Get insights into materials science with ease

    Note Explore Foundation Models for Materials with an intuitive Gradio app! Test our state-of-the-art models—SMI-TED, SELFIES-TED, MHG-GED and POS-EGNN —on your custom datasets for both classification and regression property prediction tasks. Get insights into materials science with ease