Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,25 @@
+# RAEv2-artifacts
+
+Pre-computed generation NPZs from RAEv2 stage-2 models. Each `*.npz` is
+50000 x 256 x 256 x 3 `uint8` NHWC RGB samples under key `arr_0`.
+
+| file | task | encoder | gFID |
+|---|---|---|---:|
+| `imagenet-dinov3l-k7.npz` | ImageNet 256, label-conditional | DINOv3-L k=7 (MLS last 7 layers) | 1.025 |
+
+## Usage
+
+Compute the full metric suite (`gfid`, `fdr6`, `mind6`, ...) without
+re-running distributed sampling:
+
+```bash
+hf download nanovisionx/RAEv2-artifacts --repo-type dataset \
+    --include imagenet-dinov3l-k7.npz --local-dir artifacts/
+
+EXPERIMENT_NAME=imagenet-dinov3l-k7 uv run python src/offline_eval.py \
+    --config configs/stage2/sampling/imagenet-dinov3l-k7.yaml \
+    --npz artifacts/imagenet-dinov3l-k7.npz
+```
+
+See the [main repo](https://github.com/nanovisionx/RAEv2) for the model code
+and yaml configs.