init: uploading model and architecture

README.md

@@ -1,3 +1,53 @@
----
-license: apache-2.0
----
+---
+library_name: timm
+license: apache-2.0
+datasets:
+- cifar10
+tags:
+- vision
+- image-classification
+- cifar10
+- vit
+model-index:
+- name: vit-cifar10
+  results:
+  - task: {type: image-classification}
+    dataset: {name: CIFAR-10, type: cifar10}
+    metrics:
+    - type: accuracy
+      value: 0.95   # replace with your test accuracy
+---
+
+# ViT Base (patch16, 224) fine-tuned on CIFAR-10
+
+Trained on CIFAR-10 (10 classes). Weights saved as a plain PyTorch `state_dict` (`pytorch_model.bin`).
+Architecture is defined in `vit_model.py` (uses `timm`).
+
+## Usage
+
+```python
+import torch, json
+from huggingface_hub import hf_hub_download
+import importlib.util
+
+repo_id = "roylvzn/vit-cifar10"
+
+# fetch files
+weights_path = hf_hub_download(repo_id, "pytorch_model.bin")
+model_py     = hf_hub_download(repo_id, "vit_model.py")
+classes_path = hf_hub_download(repo_id, "classes.json")
+
+# import vit_model.py dynamically
+spec = importlib.util.spec_from_file_location("vit_model", model_py)
+vm = importlib.util.module_from_spec(spec); spec.loader.exec_module(vm)
+
+# build model and load weights
+model = vm.ViTModel(num_classes=10, pretrained=False)
+state = torch.load(weights_path, map_location="cpu")
+model.load_state_dict(state)
+model.eval()
+
+with open(classes_path) as f:
+    classes = json.load(f)
+
+# inference expects 224x224 ImageNet-normalized tensors

cifar10_classes.json

@@ -0,0 +1 @@
+["airplane", "automobile", "bird", "cat", "deer", "dog", "frog", "horse", "ship", "truck"]

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cad4c8468fd4eab092a7b0c7f6c7bdd9b0bf4c337d255967e81594736b3beff2
+size 343286237

vit_model.py

@@ -0,0 +1,11 @@
+import timm
+import torch.nn as nn
+
+class ViTModel(nn.Module):
+  def __init__(self, num_classes):
+    super(ViTModel, self).__init__()
+    self.model = timm.create_model('vit_base_patch16_224', pretrained=False)
+    self.model.head = nn.Linear(self.model.head.in_features,num_classes)
+
+  def forward(self, x):
+    return self.model(x)