Avoid redundant model loading.

openshape/demo/caption.py

@@ -2,7 +2,7 @@ from torch import nn
 import numpy as np
 import torch
 from typing import Tuple, List, Union, Optional
-from transformers import GPT2Tokenizer, GPT2LMHeadModel
+from transformers import GPT2Tokenizer, GPT2LMHeadModel, GPT2Config
 from huggingface_hub import hf_hub_download
 
 
@@ -60,7 +60,7 @@ class ClipCaptionModel(nn.Module):
     def __init__(self, prefix_length: int, prefix_size: int = 512):
         super(ClipCaptionModel, self).__init__()
         self.prefix_length = prefix_length
-        self.gpt = GPT2LMHeadModel.from_pretrained('gpt2')
+        self.gpt = GPT2LMHeadModel(GPT2Config.from_pretrained('gpt2'))
         self.gpt_embedding_size = self.gpt.transformer.wte.weight.shape[1]
         if prefix_length > 10:  # not enough memory
             self.clip_project = nn.Linear(prefix_size, self.gpt_embedding_size * prefix_length)