/
Artificial Intelligence Tools and Resources

Artificial Intelligence Tools and Resources

This page provides an overview of AI tools and resources available on Sol. It includes links to technical documentation, example use-cases, and information about how to request support.

General LLM Usage

Overview:
Pretrained models from Hugging Face are centrally available in /data/datasets/community/huggingface, reducing the need for redundant downloads and making workflows faster. These models cover a wide range of NLP tasks such as classification, summarization, and question answering. This shared access enables efficient experimentation without internet access.

Specifications:

  • Additional models can be requested using the AI request form.

  • Use the path to the model when loading using the Python Transformers library from HuggingFace. Either use a pipeline or load the model directly.

Latest LLMs installed:

  • OpenAI: gpt-oss-20B

  • AstoMLab: AstroSage 8B

  • Google: Gemma 3n E4B

  • MistralAI: Magistral Small 2506

  • OpenDFM: ChemDFM 1.5 8B

  • Google: Gemma 3 12B/27B

  • Qwen: Qwen3 4B/8B/30B

  • DeepSeek-AI: DeepSeek R1 distilled Llama 8B

  • TIIAE: Falcon3 10B

Ollama

Overview:
Ollama allows you to run and experiment with open-source LLMs on Sol using containerized backends. It includes a Python API for scripting and supports a wide variety of lightweight models.

Specifications:

Use-Cases:

  • Lightweight LLM inference and chat-like interaction.

  • Prototyping model workflows with a Python backend.

AlphaFold 3

Overview:
AlphaFold 3 is a state-of-the-art protein structure prediction tool developed by DeepMind and Isomorphic Labs. It builds on earlier models by supporting complex multimer assemblies and integrating with large-scale biological datasets. ASU provides a containerized version on Sol with preconfigured settings for ease of use.

Specifications:

Use-Cases:

  • Predicting protein structures from sequence data.

  • Exploring multimeric complex predictions.

MONAI

Overview:
The Medical Open Network for AI (MONAI) is a PyTorch-based framework optimized for healthcare imaging. This tool is developed by Project MONAI, a collaboration between NVIDIA and King's College London.

Specifications:

  • GPU-enabled, installable in virtual environments.

  • Compatible with major imaging formats.

Use-Cases:

  • Deep learning in medical imaging (MRI, CT).

  • Segmentation, classification, and reconstruction.

  • Integration with PyTorch pipelines.

Biomolecular Emulator (BioEmu)

Overview:
BioEmu is a transformer-based model for learning representations of protein sequences. Developed by Microsoft, it enables downstream applications like binding prediction, variant effect prediction, and protein classification without task-specific training

Specifications:

Use-Cases:

  • Protein-ligand binding simulations.

  • AI-augmented molecular dynamics.

  • Integration into ML pipelines for biological discovery.

Evo2

Overview:
Evo2 is a generative model for protein sequences trained on evolutionary-scale datasets. Developed by the Arc Institute, it is capable of structure and function prediction across deep phylogenetic distances and can be used to explore mutational fitness landscapes.

Specifications:

  • Use the evo2 public Mamba environment on Sol.

  • Repository, including example notebooks, is downloaded in /data/sse/notebooks/evo2.

  • Documentation and usage guides are available on the project’s GitHub repository.

Use-Cases:

  • Evolutionary algorithm simulations.

  • Generative models for biology and design.

  • Phylogenetic-scale prediction tasks.

AI Scientist v2

Overview:
AI Scientist (v2) is an open-source framework for automated scientific discovery using LLM agents. Developed by Sakana AI, it orchestrates autonomous research workflows including reading literature, generating hypotheses, and simulating experiments.

Specifications:

  • Use the ai-scientist-v2 public Mamba environment on Sol.

  • Documentation and usage guides are available on the project’s GitHub repository.

  • Requires an OpenAI API key.

Use-Cases:

  • Literature synthesis and hypothesis generation.

  • Simulated research workflows.

  • Autonomous agent-based discovery.

BADASS

BADASS (Biologically-Aware Dynamic Agent Simulation System) is an LLM-based agent framework for simulating scientific interactions in biology education and research. Developed by Solugen, it enables exploratory learning, scientific inquiry, and teaching support using multiple intelligent agents.

Specifications:

  • Project notebooks are available at /data/sse/notebooks/BADASS/notebooks.

  • Source code available on GitHub.

Use-Cases:

  • Simulated classroom or lab experiments.

  • Biologically contextualized agent interactions.

  • Educational tool for teaching life sciences with AI.

For support or access to restricted tools, submit an AI support for research request.