Update app.py

app.py

@@ -6,6 +6,7 @@ from utils import load_hyperparam, load_model
 from models.tokenize import Tokenizer
 from models.llama import *
 from generate import LmGeneration
+from huggingface_hub import hf_hub_download
 
 import os
 os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
@@ -17,15 +18,16 @@ def init_args():
     global args
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     args = parser.parse_args()
-    args.load_model_path = './model_file/chatllama_7b.bin'
-    args.config_path = './config/llama_7b.json'
+    args.load_model_path = 'Linly-AI/ChatFlow-13B'
+    # args.load_model_path = './model_file/chatllama_7b.bin'
+    # args.config_path = './config/llama_7b.json'
     #args.load_model_path = './model_file/chatflow_13b.bin'
-    #args.config_path = './config/llama_13b_config.json'
+    args.config_path = './config/llama_13b_config.json'
     args.spm_model_path = './model_file/tokenizer.model'
     args.batch_size = 1
     args.seq_length = 1024
     args.world_size = 1
-    args.use_int8 = False
+    args.use_int8 = True
     args.top_p = 0
     args.repetition_penalty_range = 1024
     args.repetition_penalty_slope = 0
@@ -42,6 +44,7 @@ def init_model():
     torch.set_default_tensor_type(torch.HalfTensor)
     model = LLaMa(args)
     torch.set_default_tensor_type(torch.FloatTensor)
+    args.load_model_path = hf_hub_download(repo_id=args.load_model_path, filename='chatflow_13b.bin')
     model = load_model(model, args.load_model_path)
     model.eval()