---
license: mit
tags:
  - crystal-generation
  - diffusion-model
  - materials-science
  - probe-gradient-guidance
library_name: pytorch
---

# Crystalite 10K (Alex-MP-20)

Crystalite checkpoint trained for 10K steps on the full Alex-MP-20 dataset (540K structures, 97.9% metals). This is the diversity-optimized model used for the Pareto sweep experiments.

**Architecture**: 67.8M-parameter Diffusion Transformer with subatomic tokenizer and GEM attention bias ([Crystalite](https://arxiv.org/abs/2604.02270), Hadzi Veljkovic et al.).

## Key results with probe-gradient guidance

| Guidance weight | In-window (4-6 eV) | Uniqueness | Metal % |
|---|---|---|---|
| 0 (baseline) | 0.1% | 99.7% | 96.9% |
| 10 | 31.8% | 99.7% | 0.1% |
| 15 | 33.7% | 99.6% | 0.0% |

Every guidance weight Pareto-dominates the baseline. 18,432 structures across 6 weights, 3 seeds, 1,024 per batch. No mode collapse.

Band gap probe AUROC: 0.957 (256 parameters, trained on atom-mean hidden states).

## Usage

Requires the [Crystalite](https://github.com/joshrosie/crystalite) codebase and [probe-gradient-guidance](https://github.com/Dynamical-Systems-Research/probe-gradient-guidance) scripts.

```python
from scripts.train_probe import load_model
model = load_model("final.pt", device="cuda")
```

## Links

- **Blog post**: [Scaling Test-Time Verification for Novel Materials](https://dynamicalsystems.ai/blog/scaling-test-time-verification)
- **Code**: [Dynamical-Systems-Research/probe-gradient-guidance](https://github.com/Dynamical-Systems-Research/probe-gradient-guidance)
- **Crystalite paper**: [arXiv:2604.02270](https://arxiv.org/abs/2604.02270)

## Used In

This checkpoint was used as an upstream generation asset in the open-world environment pipeline for **Training Scientific Judgment with Verified Environments for Autonomous Science**.

- **Scientific judgment blog post**: [Training Scientific Judgment](https://dynamicalsystems.ai/blog/training-scientific-judgment)
- **Public repo**: [Dynamical-Systems-Research/training-scientific-judgment](https://github.com/Dynamical-Systems-Research/training-scientific-judgment)
- **Paper PDF**: [Training Scientific Judgment with Verified Environments for Autonomous Science](https://github.com/Dynamical-Systems-Research/training-scientific-judgment/blob/main/paper/training-scientific-judgment.pdf)