From 148f8fc000257600fa90b09e785db2733032b1cb Mon Sep 17 00:00:00 2001
From: Daniel Kaplan <dkaplan@localhost.localdomain>
Date: Wed, 3 Apr 2024 10:27:26 -0400
Subject: [PATCH] other

---
 mytests/test_vision_encoder_vit8.py | 54 +++++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)
 create mode 100644 mytests/test_vision_encoder_vit8.py

diff --git a/mytests/test_vision_encoder_vit8.py b/mytests/test_vision_encoder_vit8.py
new file mode 100644
index 000000000..1094c50f5
--- /dev/null
+++ b/mytests/test_vision_encoder_vit8.py
@@ -0,0 +1,54 @@
+import torch
+# from .vision_encoder import get_vision_encoder
+# from dinov2.models import vision_transformer as vits
+# from megatron.model.encoders.vision.vision_encoder import get_vision_encoder
+import torch
+from PIL import Image
+import open_clip
+    
+class Args:
+    def __init__(self):
+        
+        pass
+
+def test_eva_clip_image_transformer_shapes():
+    
+    device = "cuda"
+    
+    model, preprocess = open_clip.create_model_from_pretrained('ViT-B-16','hf-hub:timm/vit_base_patch16_224.augreg_in21k')
+    
+    for transform in preprocess.transforms:
+        print(transform)
+    # preprocess.transforms = preprocess.transforms[3:]
+    # print(preprocess.transforms)
+    
+    visual = model.visual
+    visual.trunk.output_tokens = True
+    
+    visual = visual.to(device)
+    del model
+    
+    image_path = "dog.jpg"
+    image = preprocess(Image.open(image_path)).unsqueeze(0).to(device)
+    print(image.shape)
+
+    pooled = visual(image)
+    
+    print(pooled.shape)
+    # print(tokens.shape)
+    
+    
+
+def main():
+    
+
+    test_eva_clip_image_transformer_shapes()
+    # test_eva_clip_image_transformer()
+    # test_dino_image_frozen_transformer()
+    # test_dino_image_frozen_lora_transformer()
+    # test_dino_video_transformer_basic()
+    
+if __name__ == "__main__":
+    
+    
+    main()