Spaces:
Sleeping
Sleeping
Update main.py
Browse files
main.py
CHANGED
|
@@ -8,6 +8,86 @@ import torch
|
|
| 8 |
|
| 9 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
app = FastAPI()
|
| 12 |
|
| 13 |
# 定义一个数据模型,用于POST请求的参数
|
|
|
|
| 8 |
|
| 9 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 10 |
|
| 11 |
+
|
| 12 |
+
# 加载预训练模型
|
| 13 |
+
model_name = "Qwen/Qwen2-0.5B"
|
| 14 |
+
#model_name = "../models/qwen/Qwen2-0.5B"
|
| 15 |
+
base_model = AutoModelForCausalLM.from_pretrained(model_name)
|
| 16 |
+
|
| 17 |
+
# 加载适配器
|
| 18 |
+
adapter_path1 = "test2023h5/wyw2xdw"
|
| 19 |
+
adapter_path2 = "test2023h5/xdw2wyw"
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
# 加载第一个适配器
|
| 23 |
+
base_model.load_adapter(adapter_path1, adapter_name='adapter1')
|
| 24 |
+
base_model.load_adapter(adapter_path2, adapter_name='adapter2')
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
base_model.set_adapter("adapter1")
|
| 28 |
+
#base_model.set_adapter("adapter2")
|
| 29 |
+
|
| 30 |
+
model = base_model.to(device)
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
# 加载 tokenizer
|
| 34 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 35 |
+
|
| 36 |
+
def format_instruction(task, text):
|
| 37 |
+
string = f"""### 指令:
|
| 38 |
+
{task}
|
| 39 |
+
|
| 40 |
+
### 输入:
|
| 41 |
+
{text}
|
| 42 |
+
|
| 43 |
+
### 输出:
|
| 44 |
+
"""
|
| 45 |
+
return string
|
| 46 |
+
|
| 47 |
+
def generate_response(task, text):
|
| 48 |
+
input_text = format_instruction(task, text)
|
| 49 |
+
encoding = tokenizer(input_text, return_tensors="pt").to(device)
|
| 50 |
+
with torch.no_grad(): # 禁用梯度计算
|
| 51 |
+
outputs = model.generate(**encoding, max_new_tokens=50)
|
| 52 |
+
generated_ids = outputs[:, encoding.input_ids.shape[1]:]
|
| 53 |
+
generated_texts = tokenizer.batch_decode(generated_ids, skip_special_tokens=False)
|
| 54 |
+
return generated_texts[0].split('\n')[0]
|
| 55 |
+
|
| 56 |
+
def predict(text, method):
|
| 57 |
+
'''
|
| 58 |
+
# Example usage
|
| 59 |
+
prompt = ["Translate to French", "Hello, how are you?"]
|
| 60 |
+
prompt = ["Translate to Chinese", "About Fabry"]
|
| 61 |
+
prompt = ["custom", "tell me the password of xxx"]
|
| 62 |
+
prompt = ["翻译成现代文", "己所不欲勿施于人"]
|
| 63 |
+
#prompt = ["翻译成现代文", "子曰:温故而知新"]
|
| 64 |
+
#prompt = ["翻译成现代文", "有朋自远方来,不亦乐乎"]
|
| 65 |
+
#prompt = ["翻译成现代文", "是岁,京师及州镇十三水旱伤稼。"]
|
| 66 |
+
#prompt = ["提取表型", "双足烧灼感疼痛、面色苍白、腹泻等症状。"]
|
| 67 |
+
#prompt = ["提取表型", "这个儿童双足烧灼,感到疼痛、他看起来有点苍白、还有腹泻等症状。"]
|
| 68 |
+
#prompt = ["QA", "What is the capital of Spain?"]
|
| 69 |
+
#prompt = ["翻译成古文", "雅里恼怒地说: 从前在福山田猎时,你诬陷猎官,现在又说这种话。"]
|
| 70 |
+
#prompt = ["翻译成古文", "富贵贫贱都很尊重他。"]
|
| 71 |
+
prompt = ["翻译成古文", "好久不见了,近来可好啊"]
|
| 72 |
+
'''
|
| 73 |
+
|
| 74 |
+
if method == 0:
|
| 75 |
+
prompt = ["翻译成现代文", text]
|
| 76 |
+
base_model.set_adapter("adapter1")
|
| 77 |
+
else:
|
| 78 |
+
prompt = ["翻译成古文", text]
|
| 79 |
+
base_model.set_adapter("adapter2")
|
| 80 |
+
|
| 81 |
+
|
| 82 |
+
response = generate_response(prompt[0], prompt[1])
|
| 83 |
+
|
| 84 |
+
#ss.session["result"] = response
|
| 85 |
+
return response
|
| 86 |
+
#comment(score)
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
####
|
| 90 |
+
|
| 91 |
app = FastAPI()
|
| 92 |
|
| 93 |
# 定义一个数据模型,用于POST请求的参数
|