指定BNB_CUDA_VERSION in bitsandbytes

app.py

@@ -2,5 +2,6 @@ import os
 
 os.system(
     f"git clone https://github.com/TimDettmers/bitsandbytes.git /home/user/app/bitsandbytes")
+os.system(f"export BNB_CUDA_VERSION=113")
 os.system(f"cd /home/user/app/bitsandbytes && CUDA_VERSION=113 make cuda11x && python setup.py install")
 os.system(f"python /home/user/app/launch.py")

launch.py

@@ -28,6 +28,7 @@ def is_chinese(text):
 AUTH_TOKEN = os.getenv("AUTH_TOKEN")
 
 LM_MODEL_PATH = "wuxiaojun/Ziya-LLaMA-13B-v1"
+# LM_MODEL_PATH = "/cognitive_comp/wuxiaojun/pretrained/pytorch/huggingface/Ziya-LLaMA-13B-v1"
 lm_model = LlamaForCausalLM.from_pretrained(
     LM_MODEL_PATH,
     device_map="auto",
@@ -36,6 +37,7 @@ lm_model = LlamaForCausalLM.from_pretrained(
     quantization_config=BitsAndBytesConfig(load_in_4bit=True))
 
 TOKENIZER_PATH = "IDEA-CCNL/Ziya-LLaMA-13B-v1"
+# TOKENIZER_PATH = "/cognitive_comp/wuxiaojun/pretrained/pytorch/huggingface/Ziya-LLaMA-13B-v1"
 # tokenizer = LlamaTokenizer.from_pretrained(LM_MODEL_PATH, use_auth_token=AUTH_TOKEN)
 tokenizer = LlamaTokenizer.from_pretrained(TOKENIZER_PATH)
 
@@ -43,8 +45,10 @@ tokenizer = LlamaTokenizer.from_pretrained(TOKENIZER_PATH)
 OPENAI_CLIP_MEAN = [0.48145466, 0.4578275, 0.40821073]
 OPENAI_CLIP_STD = [0.26862954, 0.26130258, 0.27577711]
 # demo.py is in the project path, so we can use local path ".". Otherwise you should use "IDEA-CCNL/Ziya-BLIP2-14B-Visual-v1"
+visual_model_path = "IDEA-CCNL/Ziya-BLIP2-14B-Visual-v1"
+# visual_model_path = "/cognitive_comp/wuxiaojun/pretrained/pytorch/huggingface/Ziya-BLIP2-14B-Visual-v1"
 model = AutoModelForCausalLM.from_pretrained(
-    "IDEA-CCNL/Ziya-BLIP2-14B-Visual-v1", 
+    visual_model_path,
     trust_remote_code=True, use_auth_token=AUTH_TOKEN,
     torch_dtype=torch.float16)
 model.cuda()  # if you use on cpu, comment this line