[Icicle-announce] ICICLE Release 2025-05

Mon May 5 16:45:46 EDT 2025

ICICLE Release 2025-05
The ICICLE team aims to build the next generation Cyberinfrastructure (CI) to render Artificial Intelligence (AI) more accessible to everyone and to drive its democratization further in solving larger societal problems.
It is with great pleasure that we announce 2025-05 release of ICICLE CI components.
This release includes the following components:
New to ICICLE CI Catalog
Intelligent Cyberinfrastructure
AI Foundations

  *   Forte v0.1.0<https://urldefense.com/v3/__https:/github.com/ICICLE-ai/forte-api__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9K8htFdc$>
     *   A robust out-of-distribution (OOD) detection middleware that protects ML models from processing data they weren't trained to handle. Forte uses a multi-stage approach leveraging pretrained vision models (CLIP, ViT-MSN, DINOv2) to extract features, compute topology-aware representations, and detect anomalies without requiring OOD examples during training.
  *   PEFT Vision v0.1<https://urldefense.com/v3/__https:/github.com/ICICLE-ai/PEFT_Vision__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9qv5eh54$>
     *   A Systematic Framework for Parameter-Efficient Fine Tuning (PEFT) in Visual Recognition. PEFT Vision implements 16 parameter-efficient fine-tuning (PEFT) methods, and enables consistent, reproducible evaluations of PEFT techniques across computer-vision workloads.
AI for CI-for-AI

  *   HLO Feature Dataset for AI Resource Estimation v0.0.1<https://urldefense.com/v3/__https:/huggingface.co/datasets/ICICLE-AI/ResourceEstimation_HLOGenCNN__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9n8izxqw$>
     *   A dataset containing High-Level Optimizer (HLO) graph features and metadata extracted from deep learning training workloads. It supports tasks such as runtime prediction, resource estimation, graph machine learning, and HPC scheduling optimization. Each entry includes model configuration metadata and associated HLO graph data stored in .npz format.
  *   Smart Compiler v1.0<https://urldefense.com/v3/__https:/github.com/ICICLE-ai/SMART-COMPILER__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9uOB7dao$>
     *   A Smart Compiler using AI models and traditional compiler techniques to enhance the performance scalability of C programs and Python programs. By profiling, and finding approaches for optimizations.
Software Architecture and Design

  *   ICICLEAI TapisUI Extension v0.1.0<https://urldefense.com/v3/__https:/github.com/ICICLE-ai/tapisui-extension-icicle/tree/main__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9rT-rt2Y$>
     *   A TapisUI extension written in Typescript/React for exposing ICICLEAI functionality to TapisUI deployments for the 'icicleai' tenant.
Use Inspired Science
Smart Foodsheds

  *   Food Flow Prediction GNN v1.0.0<https://urldefense.com/v3/__https:/huggingface.co/ICICLE-AI/FoodFlow_GNN_Model__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9qi23OQY$>
     *   A graph Neural Network model for predicting food flow between FAF zones using GAT and GCN architectures. The model feature include Graph Attention Networks (GAT), Graph Convolutional Networks (GCN), hurdle model for zero-inflated regression, edge feature processing, and node embedding.
ICICLE CI Components Changelog
Intelligent Cyberinfrastructure
CI-for-AI

  *   ArrayMorph For Cloud HDF5 v1.1<https://urldefense.com/v3/__https:/github.com/ICICLE-ai/ArrayMorph/tree/main__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9CD0sqIg$>
     *   Support for Microsoft Azure.
     *   File upload/download performance optimization.
  *   Hardware & Software Provisioning Service v0.2<https://urldefense.com/v3/__https:/github.com/ICICLE-ai/ct-controller/tree/main__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9z3mMlN0$>
     *   Delete the remote run directory after the application has finished running.
     *   If a specific model id is specified to the camera traps applications, pass it as model_id instead of model_type.
     *   Specify a 6 hour lease for Chameleon hardware, rather than the default 1 day.
     *   Use ad-hoc floating IPs rather than floating IP reservation leases.
Software Architecture and Design

  *   Base ICICLE Tapis Software v1.8.2<https://urldefense.com/v3/__https:/tapis.readthedocs.io/en/latest/index.html__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9ZZKNrNs$>
     *   Bug fixes.
The ICICLE team is committed to delivering the best software and CI components. We welcome your feedback and suggestions for future releases. A list of all ICICLE components can be found on our website under CI & Software<https://urldefense.com/v3/__https:/icicle.ai/cyberinfrastructure/software__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9n7Pipww$>.
Training materials for the components can be found under Education & Outreach | Training Materials<https://urldefense.com/v3/__https:/icicle-ai.github.io/training-catalog/__;!!KGKeukY!1zWtxDLMKaeZE7g-tY9kbnHqa88Hu3MNL6qRfUFJ_8aOMGVMzdr_juPxi3Ivr4BPqsRWM6li9wx4vNmG7P61X7hPGOG9FsTAIoA$>.
Please subscribe to icicle-discuss<https://lists.osu.edu/mailman/listinfo/icicle-discuss> and post to discuss all installation/build problems, performance issues, features and other miscellaneous questions related to the different software and CI components of the ICICLE project. You are welcome to post patches and enhancements to the released components.
Subscribe to our mailing list icicle-announce<https://lists.osu.edu/mailman/listinfo/icicle-announce> to stay up to date on the latest ICICLE news and releases.
Acknowledgements
This release is brought to you by the National Science Foundation (NSF) funded AI institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE) (OAC 2112606)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osu.edu/pipermail/icicle-announce/attachments/20250505/85e0d981/attachment-0001.html>