README.md

---
library_name: lucid
license: mit
tags:
  - world-modeling
  - diamond
  - lucid
datasets:
  - atari-100k
pipeline_tag: reinforcement-learning
---

# DIAMOND (CSGO)

> https://arxiv.org/abs/2405.12399

[Lucid](https://github.com/ChanLumerico/lucid) port of `https://huggingface.co/eloialonso/diamond`,
converted to Lucid-native safetensors.

## Available weights

| Tag | Params | GFLOPs | Size | Source |
|---|---|---|---|---|
| `Alien` *(default)* | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Amidar` | 0.0M | — | 51.7 MB | eloialonso/diamond |
| `Assault` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `Asterix` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `BankHeist` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `BattleZone` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Boxing` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Breakout` | 0.0M | — | 51.68 MB | eloialonso/diamond |
| `CSGO` | 0.0M | — | 1455.92 MB | eloialonso/diamond |
| `ChopperCommand` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `CrazyClimber` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `DemonAttack` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `Freeway` | 0.0M | — | 51.68 MB | eloialonso/diamond |
| `Frostbite` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Gopher` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `Hero` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Jamesbond` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Kangaroo` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Krull` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `KungFuMaster` | 0.0M | — | 51.71 MB | eloialonso/diamond |
| `MsPacman` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `Pong` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `PrivateEye` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Qbert` | 0.0M | — | 51.69 MB | eloialonso/diamond |
| `RoadRunner` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `Seaquest` | 0.0M | — | 51.72 MB | eloialonso/diamond |
| `UpNDown` | 0.0M | — | 51.69 MB | eloialonso/diamond |

## Usage

```python
import lucid.models as models
from lucid.models.weights import DiamondWeights

# default tag
model = models.diamond_csgo(pretrained=True)

# explicit tag (enum or string)
model = models.diamond_csgo(weights=DiamondWeights.CSGO)
model = models.diamond_csgo(pretrained="CSGO")

# preprocessing travels with the weights
weights = DiamondWeights.CSGO
preprocess = weights.transforms()
out = model(preprocess(image)[None])
logits = out.logits  # (B, num_classes)
```

## Conversion

Converted from `https://huggingface.co/eloialonso/diamond` via
`python -m tools.convert_weights diamond --tag CSGO`.
Key mapping + numerical parity verified against the source.

## License

`mit` — inherited from the original weights.

## Citation

```
@inproceedings{alonso2024diffusion,
  title={Diffusion for World Modeling: Visual Details Matter in Atari},
  author={Alonso, Eloi and Jelley, Adam and Micheli, Vincent and Kanervisto, Anssi and Storkey, Amos and Pearce, Tim and Fleuret, Fran{\c{c}}ois},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2024}
}
```