Perturb submission

README.md

@@ -0,0 +1,21 @@
+---
+license: apache-2.0
+library_name: torchvision
+tags: [image-classification, adversarial-training, efficientnet, perturb, bittensor]
+---
+
+# ANGExllL/rust-kestrel
+
+EfficientNetV2-L (torchvision `efficientnet_v2_l`, 1000 ImageNet classes) fine-tuned with adversarial
+training for the [Perturb](https://perturbai.io) subnet (netuid 26).
+
+- miner hotkey: `5Ebm5cRFAzkHGtq3myyvnL3w6GVSPrAy4LY6HSFMNdQLnomj`
+- on-chain hash: sha256(model.safetensors || hotkey) = `3ed9b9a728b30de319533158df890550fe866833085f7cc398b11469bcfe976a`
+- preprocessing: `EfficientNet_V2_L_Weights.IMAGENET1K_V1.transforms()` (bicubic resize 480, center crop 480, mean = std = 0.5)
+
+```python
+from safetensors.torch import load_file
+from torchvision.models import efficientnet_v2_l
+model = efficientnet_v2_l(weights=None)
+model.load_state_dict(load_file("model.safetensors"))
+```

config.json

@@ -0,0 +1,23 @@
+{
+  "architecture": "efficientnet_v2_l",
+  "framework": "torchvision",
+  "num_classes": 1000,
+  "base_weights": "torchvision/EfficientNet_V2_L_Weights.IMAGENET1K_V1",
+  "weights_file": "model.safetensors",
+  "preprocess": {
+    "resize": 480,
+    "crop": 480,
+    "interpolation": "bicubic",
+    "antialias": true,
+    "mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "std": [
+      0.5,
+      0.5,
+      0.5
+    ]
+  }
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c63d74ac5eb305bf797469bf6ede820dd27e78d290c177ad5889c2ce8d51d44
+size 476270096