Bibliographic record
Abstract
MACE 0.3.13 Release Notes 🚀 LAMMPS Integration Enhancements Added new MLIAP (Machine Learning InterAtomic Potential) interface for LAMMPS, enabling significantly improved performance and flexibility. (Documentation) Implemented CuEquivariance support in LAMMPS models for GPU acceleration. Added multi-GPU inference support for large-scale molecular dynamics simulations. Improved timing and profiling capabilities via environment variable controls. New command-line option in create_lammps_model.py to select between libtorch (legacy) and MLIAP formats: # Convert model to MLIAP format python -m mace.cli.create_lammps_model model.pt --format=mliap 🧮 Atomic Stresses Computation Added support for computing atomic stresses and atomic virials. Useful for analyzing local stress distributions in materials simulations. from mace.calculators import mace_mp from ase import build # Create structure with 10 atoms atoms = build.bulk("Al", "fcc", a=4.05, cubic=True) atoms = atoms.repeat((2, 2, 2)) calc = mace_mp(device="cpu", compute_atomic_stresses=True) atoms.set_calculator(calc) atoms.get_potential_energy() stress = atoms.get_stress() stresses = atoms.get_stresses() print("Stress tensor:\n", stress.shape) print("Stresses tensor:\n", stresses.shape) 🗝️ Property Keys System Redesign Reworked the property keys system for improved flexibility and maintainability. Introduced KeySpecification class to manage mappings between data formats. Added DefaultKeys enum to standardize access patterns. Enhanced error reporting when keys are missing. Fully backward compatible with previous key conventions. 🧱 New Foundation Models MACE-MATPES: New foundation models finetuned on matpes dataset. New PBE model with +U inconsistencies and a new R2SCAN model. Better transferability for fine-tuning on domain-specific datasets. | Model | Method | Energy (meV/atom) | Force (meV/Å) | Stress (GPa) | |------------------|--------|-------------------|----------------|--------------| | M3GNet | PBE | 45 | 177 | 0.898 | | CHGNet | PBE | 32 | 124 | 0.617 | | TensorNet | PBE | 36 | 138 | 0.695 | | MACE | PBE | 34 | 122 | 0.296 | | MACE-MATPES-0 | PBE | 23 | 107 | 0.304 | | M3GNet | r2SCAN | 45 | 208 | 0.982 | | CHGNet | r2SCAN | 27 | 150 | 0.705 | | TensorNet | r2SCAN | 34 | 163 | 0.754 | | MACE-MAPTES-0 | r2SCAN | 19 | 119 | 0.265 | 🧠 Improved Model Head Selection Added explicit head selection in the MACE calculator for multi-head models. You can now directly specify the head: # Example of selecting a specific head calc = MACECalculator(model_path="model.pt", head="DFT") Automatically falls back to "default" head if not specified. Clear error messages when requested head is unavailable. Head-specific configurations supported during inference. 🧪 MACE Fine-Tuning Preselection CLI MACE a tool for selecting configurations when fine-tuning foundation models. 🔍 Key Features Multiple Filtering Strategies: combinations: Only elements in your subset exclusive: Exactly your elements inclusive: All your elements plus potentially others Selection Methods: fps (Farthest Point Sampling) for maximum diversity random for uniform random sampling Weighting Control: Adjust the importance of pretraining vs. fine-tuning data during selection 💡 Usage Example python -m mace.cli.fine_tuning_select \ --configs_pt path/to/pretraining_data.xyz \ --atomic_numbers "[1, 6, 8]" \ --num_samples 5000 \ --filtering_type combinations \ --output selected_configs.xyz This functionality is also available directly in run_train.py when using the --atomic_numbers parameter. One can no longer do the filtering from run_train without adding the atomic numbers. 📦 Installation pip install --upgrade mace-torch # For CUDA acceleration (CUDA 12) pip install cuequivariance cuequivariance-torch cuequivariance-ops-torch-cu12 # For CUDA 11 pip install cuequivariance-ops-torch-cu11 For complete usage and documentation, please see our official documentation. If you encounter any issues, please report them on our GitHub Issues page. 🔄 Full Changelog Full Changelog: https://github.com/ACEsuit/mace/compare/v0.3.12...v0.3.13
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.244 | 0.055 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".