Fix: Add position_ids to CLIP encoder and config.json to match Volkopat/DeepSeek-DeepEncoder exactly

clip_encoder.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d3f97c24bd69378a5f5a657ad81223134025ebf52f784eb32042d4d2b57404f
-size 606449932
+oid sha256:2d9fb61233775d3f81587378737cc307265356fe9ddb677aa67c9e418b09872c
+size 606452303

config.json

@@ -0,0 +1,29 @@
+{
+  "sam": {
+    "params": 95569152,
+    "architecture": "SAM ViT-B",
+    "image_size": 1024,
+    "patch_size": 16,
+    "embed_dim": 768,
+    "depth": 12,
+    "num_heads": 12
+  },
+  "clip": {
+    "params": 303177728,
+    "architecture": "CLIP-Large",
+    "image_size": 224,
+    "patch_size": 14,
+    "width": 1024,
+    "layers": 24,
+    "heads": 16
+  },
+  "projector": {
+    "params": 2622720,
+    "type": "linear",
+    "input_dim": 2048,
+    "output_dim": 1280
+  },
+  "total_params": 401369600,
+  "output_tokens": 256,
+  "output_dim": 1280
+}