#4646th of 60 in Training and fine-tuningFor science, health and research, science
the_well
PyTorch datasets and benchmarks of physics simulations for PDE surrogate models
Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 4.5k
- License
- BSD-3-Clause
- Last commit
- Oct 2026
- Last release
- Nov 2025
- Language
- Jupyter Notebook
Missed the latest GitHub check; these facts may be older.
Overview
The Well is a collection of 16 numerical simulation datasets, 15TB in total, covering fluid dynamics, acoustic scattering, biological systems and astrophysical MHD. The package provides a PyTorch WellDataset class, a download CLI, Hugging Face streaming, and a Hydra-configured training script with baseline models such as FNO and pretrained checkpoints.
Who it is for: ML researchers training surrogate models for PDEs and physical simulations
Strengths
- 16 datasets from domain scientists, each between 6.9GB and 5.1TB
- WellDataset plugs into a standard PyTorch DataLoader
- Download with the-well-download or stream from Hugging Face
- Hydra configs and pretrained baseline checkpoints for reproducing the benchmark
Weaknesses
- Full collection needs 15TB of disk; streaming is slower for large training
- Benchmark models are simple baselines, not state-of-the-art, per the README
- viscoelastic_instability is deprecated for processing errors; use the v2 dataset
- Limited to physics simulation data, not a general-purpose ML library
What it needs
- GPU optional
- Needs Python >=3.10, PyTorch, Hydra, Hugging Face Hub
- Models: FNO
Also in Training and fine-tuning
See all 60| Rank | Project | Score |
|---|---|---|
| 1 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 89 out of 100 |
| 2 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 3 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 87 out of 100 |
| #4 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 86 out of 100 |
| #5 | megatron-lmDistributed training library for large transformer models on NVIDIA GPUs | 85 out of 100 |