Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -39,93 +39,44 @@ def generate_response(instruction, input_text):
|
|
39 |
global model, tokenizer
|
40 |
|
41 |
if model is None:
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
#
|
71 |
-
base_model_name = "larry1129/meta-llama-3.1-8b-bnb-4bit"
|
72 |
-
|
73 |
-
# 定义 adapter 模型名称
|
74 |
-
adapter_model_name = "larry1129/WooWoof_AI"
|
75 |
-
|
76 |
-
# 定义全局变量用于缓存模型和分词器
|
77 |
-
model = None
|
78 |
-
tokenizer = None
|
79 |
-
|
80 |
-
# 定义提示生成函数
|
81 |
-
def generate_prompt(instruction, input_text=""):
|
82 |
-
if input_text:
|
83 |
-
prompt = f"""### Instruction:
|
84 |
-
{instruction}
|
85 |
-
### Input:
|
86 |
-
{input_text}
|
87 |
-
### Response:
|
88 |
-
"""
|
89 |
else:
|
90 |
-
prompt = f"""### Instruction:
|
91 |
-
{instruction}
|
92 |
-
### Response:
|
93 |
-
"""
|
94 |
-
return prompt
|
95 |
-
|
96 |
-
# 定义生成响应的函数,并使用 @spaces.GPU 装饰
|
97 |
-
@spaces.GPU(duration=120)
|
98 |
-
def generate_response(instruction, input_text):
|
99 |
-
global model, tokenizer
|
100 |
-
|
101 |
-
if model is None:
|
102 |
-
# 检查 bitsandbytes 是否已安装
|
103 |
-
import importlib.util
|
104 |
-
if importlib.util.find_spec("bitsandbytes") is None:
|
105 |
-
import subprocess
|
106 |
-
subprocess.call(["pip", "install", "--upgrade", "bitsandbytes"])
|
107 |
-
|
108 |
-
# 在函数内部导入需要 GPU 的库
|
109 |
-
import torch
|
110 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
|
111 |
-
from peft import PeftModel
|
112 |
-
|
113 |
-
# 创建量化配置
|
114 |
-
bnb_config = BitsAndBytesConfig(
|
115 |
-
load_in_4bit=True,
|
116 |
-
bnb_4bit_use_double_quant=True,
|
117 |
-
bnb_4bit_quant_type="nf4",
|
118 |
-
bnb_4bit_compute_dtype=torch.float16
|
119 |
-
)
|
120 |
-
|
121 |
-
# 加载分词器
|
122 |
-
tokenizer = AutoTokenizer.from_pretrained(base_model_name, use_auth_token=hf_token)
|
123 |
-
|
124 |
-
# 加载基础模型
|
125 |
-
|
126 |
# 在函数内部导入需要的库
|
127 |
import torch
|
128 |
|
|
|
|
|
|
|
|
|
|
|
129 |
# 生成提示
|
130 |
prompt = generate_prompt(instruction, input_text)
|
131 |
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
|
|
39 |
global model, tokenizer
|
40 |
|
41 |
if model is None:
|
42 |
+
print("开始加载模型...")
|
43 |
+
try:
|
44 |
+
# 在函数内部导入需要 GPU 的库
|
45 |
+
import torch
|
46 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
47 |
+
|
48 |
+
# 加载分词器
|
49 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=hf_token)
|
50 |
+
print("分词器加载成功。")
|
51 |
+
|
52 |
+
# 加载模型
|
53 |
+
model = AutoModelForCausalLM.from_pretrained(
|
54 |
+
model_name,
|
55 |
+
device_map="auto",
|
56 |
+
torch_dtype=torch.float16,
|
57 |
+
use_auth_token=hf_token,
|
58 |
+
)
|
59 |
+
print("模型加载成功。")
|
60 |
+
|
61 |
+
# 设置 pad_token
|
62 |
+
tokenizer.pad_token = tokenizer.eos_token
|
63 |
+
model.config.pad_token_id = tokenizer.pad_token_id
|
64 |
+
|
65 |
+
# 切换到评估模式
|
66 |
+
model.eval()
|
67 |
+
print("模型已切换到评估模式。")
|
68 |
+
except Exception as e:
|
69 |
+
print("加载模型时出错:", e)
|
70 |
+
raise e # 重新抛出异常,以便在日志中看到完整的错误信息
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
else:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
# 在函数内部导入需要的库
|
73 |
import torch
|
74 |
|
75 |
+
# 检查 model 和 tokenizer 是否已正确加载
|
76 |
+
if model is None or tokenizer is None:
|
77 |
+
print("模型或分词器未正确加载。")
|
78 |
+
raise ValueError("模型或分词器未正确加载。")
|
79 |
+
|
80 |
# 生成提示
|
81 |
prompt = generate_prompt(instruction, input_text)
|
82 |
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|