Upload EVPRefer_warp

config.json

@@ -1,10 +1,10 @@
 {
   "architectures": [
-    "EVPRefer"
+    "EVPRefer_warp"
   ],
   "auto_map": {
     "AutoConfig": "evpconfig.EVPConfig",
-    "AutoModel": "model.EVPRefer"
+    "AutoModel": "model.EVPRefer_warp"
   },
   "model_type": "EVP",
   "torch_dtype": "float32",

model.py

@@ -243,23 +243,31 @@ class InverseMultiAttentiveFeatureRefinement(nn.Module):
         '''
         return [x_c4, x_c3, x_c2, x_c1]
         
+
+class EVPRefer_warp(PreTrainedModel):
+    config_class = EVPConfig
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = EVPRefer()
+    def forward(self, img, sentences):
+        return self.model(img, sentences)
+
         
-        
-class EVPRefer(PreTrainedModel):
+class EVPRefer(nn.Module):
     """Encoder Decoder segmentors.
 
     EncoderDecoder typically consists of backbone, decode_head, auxiliary_head.
     Note that auxiliary_head is only used for deep supervision during training,
     which could be dumped during inference.
     """
-    config_class = EVPConfig
-    def __init__(self, config,
+    #config_class = EVPConfig
+    def __init__(self, #config,
                  sd_path=None,
                  base_size=512,
                  token_embed_dim=768,
                  neck_dim=[320,680,1320,1280],
                  **args):
-        super().__init__(config)
+        super().__init__()
         config = OmegaConf.load('./v1-inference.yaml')
         if os.path.exists(f'{sd_path}'):
             config.model.params.ckpt_path = f'{sd_path}'

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:657dc9c51ec9bbcb561120186fe0f02e9381fc789cf9bb70f5006bceba8b1377
-size 4317946624
+oid sha256:8cb0158d61ae68cf82b2a478c35521123c8f7e89af87888349c97363392824a0
+size 4317953152