Update app.py

app.py

@@ -9,12 +9,12 @@ import copy
 import secrets
 from pathlib import Path
 
-tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen-VL-Chat-Int4", trust_remote_code=True)
-config = AutoConfig.from_pretrained("Qwen/Qwen-VL-Chat-Int4", trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen-VL-Chat-Int4")
+config = AutoConfig.from_pretrained("Qwen/Qwen-VL-Chat-Int4", trust_remote_code=True, torch_dtype=torch.float16)
 #config.quantization_config["use_exllama"] = True
 config.quantization_config["disable_exllama"] = False
 config.quantization_config["exllama_config"] = {"version":2}
-model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen-VL-Chat-Int4", device_map="auto", config=config).eval()
+model = AutoModelForCausalLM.from_config(config, trust_remote_code=True, torch_dtype=torch.float16)
 
 BOX_TAG_PATTERN = r"<box>([\s\S]*?)</box>"
 PUNCTUATION = "！？。＂＃＄％＆＇（）＊＋，－／：；＜＝＞＠［＼］＾＿｀｛｜｝～｟｠｢｣､、〃》「」『』【】〔〕〖〗〘〙〚〛〜〝〞〟〰〾〿–—‘’‛“”„‟…‧﹏."