Update src/models/tokenizer/qwen_visual.py

src/models/tokenizer/qwen_visual.py

@@ -18,7 +18,7 @@ from torch.nn import functional as F
 from torch.nn.init import trunc_normal_
 from torchvision import transforms
 from torchvision.transforms import InterpolationMode
-from src.models.mllm.utils import load_zero3_checkpoint
+#from src.models.mllm.utils import load_zero3_checkpoint
 
 
 def get_abs_pos(abs_pos, tgt_size):
@@ -450,8 +450,8 @@ class VisionTransformerWithAttnPool(nn.Module):
         if pretrained_model_path is not None:
             print('Load ckpt of qwen visual encoder')
             ckpt = torch.load(pretrained_model_path, map_location='cpu')
-            # missing, unexpected = model.load_state_dict(ckpt, strict=False)
-            load_zero3_checkpoint(model, ckpt)
+            missing, unexpected = model.load_state_dict(ckpt, strict=False)
+            #load_zero3_checkpoint(model, ckpt)
             # print('Load ckpt of qwen visual encoder')
             # print('missing keys: ', len(missing), 'unexpected keys:', len(unexpected))