Switched to HF automodel.

app.py

@@ -18,9 +18,9 @@ T4级别的GPU已经可以很胜任这个任务了。
 import torch
 import sys
 
-from transformers import AutoTokenizer, GenerationConfig, AutoModelForCausalLM
+from transformers import AutoTokenizer, GenerationConfig, AutoModel
 
-model = AutoModelForCausalLM.from_pretrained("ljsabc/Fujisaki-int4", trust_remote_code=True).float()
+model = AutoModel.from_pretrained("ljsabc/Fujisaki-int4", trust_remote_code=True).float()
 tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True)
 
 # We have to use full precision, as some tokens are >65535