Update modeling_internvideo2.py

modeling_internvideo2.py

@@ -1066,10 +1066,10 @@ def pretrain_internvideo2_6b_patch14_224(config):
         embed_dim=3200, depth=48, num_heads=25, mlp_ratio=4,
         clip_embed_dim=config.vision_encoder.clip_embed_dim,
         attn_pool_num_heads=16, qkv_bias=False,
-        # drop_path_rate=0.3,
-        # init_values=0.00001,
         drop_path_rate=0,
-        init_values=None,
+        init_values=0.00001,
+        # drop_path_rate=0,
+        # init_values=None,
         qk_normalization=True,
         use_flash_attn=config.vision_encoder.use_flash_attn,
         use_fused_rmsnorm=config.vision_encoder.use_fused_rmsnorm,