diff --git a/.claude/commands/add-model.md b/.claude/commands/add-model.md index 68f9221..440893c 100644 --- a/.claude/commands/add-model.md +++ b/.claude/commands/add-model.md @@ -18,6 +18,12 @@ Add a new model to this repository following all repo conventions. - Task description - License (SPDX id or direct link) - Upstream code repo and paper links + - **Acknowledgements and citation section** (required — see template): + - Upstream repo URL + - Paper citation (author, title, venue, DOI or arXiv id) + - BibTeX or DOI-based "cite as" block copied from the upstream repo or model card + - ROCm blog URL and author names (AMD Silo AI) if a blog post exists + - Any named collaboration (e.g. AstraZeneca × AMD, ORNL × AMD) 5. **Add recipes** under `/recipes/`. Use one subfolder per task (`recipes/inference/`, `recipes/finetune/`, etc.). Each recipe subfolder needs a `README.md`. @@ -42,9 +48,11 @@ Add a new model to this repository following all repo conventions. - `healthcare/` (Healthcare & Life Sciences): add research/engineering-only disclaimer; no PHI; copy intended-use and limitations from the model card. - `physics_simulation/`: state physical domain (fluid dynamics, plasma, etc.), dataset format (HDF5, NetCDF), and HPC/multi-node requirements. -9. **Do not commit** large checkpoints, datasets, `.env` files, or tokens — document how users obtain them instead. +9. **Update `ACKNOWLEDGEMENTS.md`** at the repo root — add a per-model entry under the appropriate domain section following the existing format. Include paper citation, upstream repo, ROCm blog + author (if applicable), and any collaboration callout. -10. **Git workflow** — always branch, never commit directly to `main`: +10. **Do not commit** large checkpoints, datasets, `.env` files, or tokens — document how users obtain them instead. + +11. **Git workflow** — always branch, never commit directly to `main`: 1. `git fetch origin && git checkout -b / origin/main` 2. Create all files, set scripts `chmod +x`. 3. `git add /models//` and commit. diff --git a/ACKNOWLEDGEMENTS.md b/ACKNOWLEDGEMENTS.md new file mode 100644 index 0000000..fb203c8 --- /dev/null +++ b/ACKNOWLEDGEMENTS.md @@ -0,0 +1,112 @@ +# Acknowledgements + +AI4Science Studio curates recipes for open AI-for-science models. The models themselves, their weights, and the ideas behind them are the work of the original authors listed below. AMD Silo AI contributed ROCm/AMD-specific recipes and blog posts for many of these models. + +--- + +## Earth Science + +### StormCast +- **Upstream:** NVIDIA Earth-2 Studio — [`NVIDIA/earth2studio`](https://github.com/NVIDIA/earth2studio) +- **Paper:** Bodnar et al., *Kilometer-Scale Convection Allowing Model Emulation using Generative Diffusion Modeling*, arXiv:2408.10958 +- **ROCm blog (inference):** [Running StormCast on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/stormcast-inference/README.html) — Pauli Pihajoki (AMD Silo AI) +- **ROCm blog (ensembles):** [StormCast Ensemble Forecasting on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/stormcast-ensembles/README.html) — Pauli Pihajoki (AMD Silo AI) + +### ORBIT-2 +- **Upstream:** [`XiaoWang-Github/ORBIT-2`](https://github.com/XiaoWang-Github/ORBIT-2); weights on [Hugging Face](https://huggingface.co/jychoi-hpc/ORBIT-2) +- **Paper:** Wang et al., *ORBIT-2: Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling*, arXiv:2505.04802 +- **Dataset DOI:** [10.13139/OLCF/2589526](https://doi.org/10.13139/OLCF/2589526) +- **Maintained by:** Oak Ridge National Laboratory (ORNL) + +### ArchesWeather +- **Upstream:** [`gcouairon/ArchesWeather`](https://huggingface.co/gcouairon/ArchesWeather); recipe code from [`silogen/ai-samples`](https://github.com/silogen/ai-samples) +- **Paper:** Couairon et al., *ArchesWeather & ArchesWeatherGen: efficient AI weather forecasting*, arXiv:2412.12971 +- **ROCm blog:** [Training ArchesWeather on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/geoarches-training/README.html) — Luka Tsabadze, Rahul Biswas, Pauli Pihajoki, Daniel Warna, Baiqiang Xia, Sopiko Kurdadze (AMD Silo AI) + +### Aurora +- **Upstream:** [`microsoft/aurora`](https://github.com/microsoft/aurora); recipe code from [`silogen/ai-samples`](https://github.com/silogen/ai-samples) +- **Paper:** Bodnar et al., *A foundation model of the Earth system*, Nature 2025, https://doi.org/10.1038/s41586-025-08897-0 +- **ROCm blog:** [Running SOTA AI-based Weather Forecasting models on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/ai-weather-forecasting/README.html) — Luka Tsabadze, Rahul Biswas, Pauli Pihajoki, Daniel Warna, Baiqiang Xia (AMD Silo AI) + +### GenCast +- **Upstream:** [`google-deepmind/graphcast`](https://github.com/google-deepmind/graphcast); recipe code from [`silogen/ai-samples`](https://github.com/silogen/ai-samples) +- **Paper:** Price et al., *GenCast: Diffusion-based ensemble weather forecasting for improved prediction accuracy and uncertainty quantification*, Nature 2025, https://doi.org/10.1038/s41586-024-08252-9 +- **ROCm blog:** [Running SOTA AI-based Weather Forecasting models on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/ai-weather-forecasting/README.html) — Luka Tsabadze, Rahul Biswas, Pauli Pihajoki, Daniel Warna, Baiqiang Xia (AMD Silo AI) + +### NeuralGCM +- **Upstream:** [`google-research/neuralgcm`](https://github.com/google-research/neuralgcm); weights via Google Cloud Storage +- **Papers:** + - Kochkov et al., *Neural General Circulation Models for Weather and Climate*, arXiv:2311.07222 + - Kochkov et al., *Neural GCMs optimized to predict satellite-based precipitation*, arXiv:2412.11973 + +### PanguWeather +- **Upstream:** [`198808xc/Pangu-Weather`](https://github.com/198808xc/Pangu-Weather); recipe code from [`silogen/ai-samples`](https://github.com/silogen/ai-samples) +- **Paper:** Bi et al., *Accurate medium-range global weather forecasting with 3D neural networks*, Nature 2023, https://doi.org/10.1038/s41586-023-06185-3 +- **ROCm blog:** [Running SOTA AI-based Weather Forecasting models on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/ai-weather-forecasting/README.html) — Luka Tsabadze, Rahul Biswas, Pauli Pihajoki, Daniel Warna, Baiqiang Xia (AMD Silo AI) + +--- + +## Material Science + +### MatterGen +- **Upstream:** [`microsoft/mattergen`](https://github.com/microsoft/mattergen) +- **Paper:** Zeni et al., *MatterGen: a generative model for inorganic materials design*, Nature 2025, https://doi.org/10.1038/s41586-025-08628-5 +- **ROCm blog:** [Running MatterGen on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/mattergen/README.html) — Sopiko Kurdadze (AMD Silo AI) + +### HydraGNN +- **Upstream:** [`ORNL/HydraGNN`](https://github.com/ORNL/HydraGNN) (branch `Predictive_GFM_2024`); weights on [Hugging Face](https://huggingface.co/mlupopa/HydraGNN_Predictive_GFM_2024) +- **Dataset DOI:** [10.13139/OLCF/2474799](https://doi.org/10.13139/OLCF/2474799) +- **Citation:** M. Lupo Pasini et al., *HydraGNN_Predictive_GFM_2024 — Ensemble of predictive graph foundation models for group state atomistic materials modeling*, DOI 10.13139/OLCF/2474799 +- **Maintained by:** Oak Ridge National Laboratory (ORNL) + +--- + +## Healthcare & Life Sciences + +### GP-MoLFormer +- **Upstream:** [`IBM/gp-molformer`](https://github.com/IBM/gp-molformer) +- **Paper:** Ross et al., *Generative Pre-trained Transformer for De Novo Drug Design and Molecular Property Optimization*, arXiv:2302.07432 +- **ROCm blog:** [Running GP-MoLFormer on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/gp-molformer/README.html) — Sopiko Kurdadze (AMD Silo AI) + +### SwinUNETR +- **Upstream:** [`Project-MONAI/research-contributions`](https://github.com/Project-MONAI/research-contributions); recipe code from [`silogen/ai-samples`](https://github.com/silogen/ai-samples) +- **Paper:** Hatamizadeh et al., *Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images*, arXiv:2201.01266 +- **ROCm blog (training):** [Running SwinUNETR on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/running-swinunetr-amd/README.html) — Joaquin Rives Gambin (AMD Silo AI) +- **ROCm blog (inference optimization):** [SwinUNETR Inference Optimization on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/swinunetr-inference-optimization/README.html) — Joaquin Rives Gambin, Vasumathi Neralla, David Björelind, Rui Sampaio (AMD Silo AI / AstraZeneca × AMD collaboration) +- **AstraZeneca × AMD collaboration:** https://www.amd.com/en/blogs/2025/astrazeneca-improved-life-sciences-model-training-time.html + +### SemlaFlow +- **Upstream:** [`rssrwn/semla-flow`](https://github.com/rssrwn/semla-flow) +- **Paper:** Morehead et al., *SemlaFlow: Efficient 3D Molecular Generation with Latent Attention and Equivariant Flow Matching*, OpenReview: https://openreview.net/forum?id=bee2G6pEh0 +- **ROCm blog:** [Running SemlaFlow on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/semlaflow/README.html) — Vasumathi Neralla, Rui Sampaio (AMD Silo AI / AstraZeneca × AMD collaboration) +- **AstraZeneca × AMD collaboration:** https://www.amd.com/en/blogs/2025/astrazeneca-improved-life-sciences-model-training-time.html + +### REINVENT4 +- **Upstream:** [`MolecularAI/REINVENT4`](https://github.com/MolecularAI/REINVENT4) +- **Paper:** Loeffler et al., *REINVENT4: Modern AI-driven generative molecule design*, J Cheminformatics 2024, https://doi.org/10.1186/s13321-024-00812-5 +- **ROCm blog:** [Running REINVENT4 on AMD Instinct](https://rocm.blogs.amd.com/artificial-intelligence/running-reinvent4-amd/README.html) — David Björelind, Rui Sampaio (AMD Silo AI / AstraZeneca × AMD collaboration) +- **AstraZeneca × AMD collaboration:** https://www.amd.com/en/blogs/2025/astrazeneca-improved-life-sciences-model-training-time.html + +--- + +## Physics Simulation + +### MATEY +- **Upstream:** [`ORNL/MATEY`](https://github.com/ORNL/MATEY) +- **Paper:** Subramanian et al., *MATEY: multiscale adaptive foundation models for spatiotemporal physical systems*, arXiv:2412.20601 +- **Maintained by:** Oak Ridge National Laboratory (ORNL) + +### Walrus +- **Upstream:** [`PolymathicAI/walrus`](https://github.com/PolymathicAI/walrus); weights on [Hugging Face](https://huggingface.co/polymathic-ai/walrus) +- **Paper:** McCabe et al., *Walrus: A Cross-Domain Foundation Model for Continuum Dynamics*, arXiv:2511.15684 +- **Maintained by:** Polymathic AI + +--- + +## AMD Silo AI + +ROCm-specific recipes for many models in this repository were authored by the [AMD Silo AI](https://www.amd.com/en/solutions/ai/silo-ai.html) team and published on the [ROCm Blogs](https://rocm.blogs.amd.com) platform. Contributors across these recipes include: Pauli Pihajoki, Luka Tsabadze, Rahul Biswas, Sopiko Kurdadze, Daniel Warna, Baiqiang Xia, Joaquin Rives Gambin, Vasumathi Neralla, David Björelind, and Rui Sampaio. + +## AstraZeneca × AMD Collaboration + +The recipes for REINVENT4, SemlaFlow, and the SwinUNETR inference optimization were developed as part of a collaboration between AstraZeneca and AMD. See the [collaboration blog post](https://www.amd.com/en/blogs/2025/astrazeneca-improved-life-sciences-model-training-time.html) for background. The SwinUNETR training recipe and GP-MoLFormer are AMD Silo AI contributions independent of this collaboration. diff --git a/_template/README.md b/_template/README.md index 0eef295..dfc3d72 100755 --- a/_template/README.md +++ b/_template/README.md @@ -16,3 +16,17 @@ Rename the parent folder from `_template` to your **model slug** (see the domain Put training, fine-tuning, inference, or evaluation scripts (or step-by-step docs) under `recipes/`. Prefer one subfolder per task, e.g. `recipes/inference/`, `recipes/finetune/`. Do not commit large checkpoints or datasets; document how to obtain them instead. + +## Acknowledgements and citation + +Every model entry must include attribution. Fill in this section before opening a PR: + +``` +- **Upstream repo:** [org/repo](https://github.com/org/repo) +- **Paper:** Author et al., *Title*, Venue Year, https://doi.org/... (or arXiv:XXXX.XXXXX) +- **Cite as:** paste BibTeX or DOI-based citation from the upstream repo/model card +- **ROCm blog (if applicable):** [Title](https://rocm.blogs.amd.com/...) — Author Name (AMD Silo AI) +- **Collaboration (if applicable):** e.g. AstraZeneca × AMD, ORNL × AMD +``` + +Also add an entry to [`ACKNOWLEDGEMENTS.md`](../../ACKNOWLEDGEMENTS.md) at the repo root following the existing per-model format.