AI Models

The UFIT Research Computing AI Support Team maintains a suite of commonly used AI models on HiPerGator. Users may copy these models to their own space, add modifications, and follow the instructions to run these jobs on HiPerGator. Each model has a readme file with additional information in its directory. Use https://support.rc.ufl.edu to submit tickets if you need help with the models or have any AI questions.

Name	Categories	Location on HiPerGator	Dataset size (approximate)	Version	License	Date added	Description
	Computer vision	/data/ai/models/computer_vision/ultralytics_yolov8	605.2 MiB	v8		5-May-24	YOLOv8 Detect, Segment and Pose models pretrained on the COCO dataset are available here, as well as YOLOv8 Classify models pretrained on the ImageNet dataset. Track mode is available for all Detect, Segment and Pose models.
	Healthcare and life science	/data/ai/models/healthcare_life_science/proteinfolding/alphafold	8.7 GiB	v2.0.0		6-Jul-22	Predicts protein structures. If you publish research using alphafold, the original paper must be cited.
	Healthcare and life science	/data/ai/models/healthcare_life_science/proteinfolding/rosettafold	1.0 GiB	v1.0.0		11-Mar-21	Predicts protein structures. If you publish research using RoseTTAFold, the original paper must be cited https://www.biorxiv.org/content/10.1101/2021.06.14.448402v
	Imaging	/data/ai/models/nvidia/stylegan3	7.2 GiB	3		29-Apr-24	StyleGAN3 is a cutting-edge generative model for high-quality image synthesis, offering unparalleled control over image style and content, making it ideal for creative and enterprise applications.
	Multimodal	/data/ai/models/multimodel/clip/clip-vit-base-patch32	3.4 GiB	openai/clip-vit-base-patch32		17-Jul-23	The clip-vit-base-patch32 uses a ViT-B/32 Transformer architecture as an image encoder and uses a masked self-attention Transformer as a text encoder. These encoders are trained to maximize the similarity of (image, text) pairs via a contrastive loss.
	Multimodal	/data/ai/models/multimodel/clip/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224	1.5 GiB	microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224		17-Jul-23	BiomedCLIP is a biomedical vision-language foundation model that is pretrained on PMC-15M, a dataset of 15 million figure-caption pairs extracted from biomedical research articles in PubMed Central, using contrastive learning. It uses PubMedBERT as the text encoder and Vision Transformer as the image encoder, with domain-specific adaptations. It can perform various vision-language processing (VLP) tasks such as cross-modal retrieval, image classification, and visual question answering.
	NLP	/data/ai/models/nlp/gemma	250.1 GiB			9-Apr-24	Gemma is a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. Developed by Google DeepMind and other teams across Google, Gemma is named after the Latin gemma, meaning "precious stone."
	NLP	/data/ai/models/nlp/llama	6.7 TiB	Llama2, Llama3		19-Apr-24	LLaMA models are powerful language models developed by Meta AI, with the latest version being LLaMA 3, which significantly improves performance and accessibility for various natural language processing tasks.
	NLP	/data/ai/models/nlp/meditron	564.2 GiB			3-May-24	Meditron is a suite of open-source medical Large Language Models (LLMs). The team provide Meditron-7B and Meditron-70B, fine-tuned for medical tasks using a diverse medical dataset. Among these, Meditron-70B shows superior performance compared to other models like Llama-2-70B, GPT-3.5, and Flan-PaLM across multiple medical reasoning tasks.
	NLP	/data/ai/models/nlp/megatron	22.3 GiB	2.2, 2.5, 3.0			Megatron-LM, a fascinating language model developed by the Applied Deep Learning Research team at NVIDIA.
	NLP	/data/ai/models/nlp/mistral_ai	875.6 GiB			9-Apr-24	Mistral AI offers a variety of language models, including open-weights models like Mistral 7B, Mixtral 8x7B, and Mixtral 8x22B, as well as optimized commercial models such as Mistral Small, Mistral Medium, Mistral Large, and Mistral Embeddings
	NLP	/data/ai/models/nvidia/bionemo/dnabert	64.3 GiB	1.2		28-Feb-24	DNABERT generates a dense representation of a genome sequence by identifying contextually similar sequences in the human genome. DNABert is a DNA sequence model trained on sequences from the human reference genome Hg38.p13.
	NLP	/data/ai/models/nvidia/nemo/nemo_24.01.gemma	21.4 GiB			18-Apr-24	NeMo framework container with the pre-trained model Gemma.
	NLP	/data/ai/models/nvidia/nemo/nemo_24.01.starcoder2	22.6 GiB	2		18-Apr-24	NeMo framework container with the pre-trained model StarCoder2.
	NLP	/data/ai/models/nvidia/nemo/nemo_24.03.codegemma	20.2 GiB			18-Apr-24	NeMo framework container with the pre-trained model CodeGemma.
	NLP	/data/ai/models/nlp/gatortron	50.2 GiB			3-May-24	GatorTron is a large clinical language model developed by researchers at the University of Florida Health in collaboration with NVIDIA. It’s designed to accelerate research and medical decision-making by extracting insights from massive volumes of clinical data with unprecedented speed and clarity.