Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -12,7 +12,7 @@ device = torch.device("cuda:{}".format(device_ids_parallel[0]) if USE_CUDA else
|
|
12 |
# 初始化
|
13 |
peft_model_id = "CMLM/ZhongJing-2-1_8b"
|
14 |
base_model_id = "Qwen/Qwen1.5-1.8B-Chat"
|
15 |
-
model = AutoModelForCausalLM.from_pretrained(base_model_id, device_map="auto")
|
16 |
model.load_adapter(peft_model_id)
|
17 |
tokenizer = AutoTokenizer.from_pretrained(
|
18 |
"CMLM/ZhongJing-2-1_8b",
|
@@ -29,8 +29,8 @@ def single_turn_chat(question):
|
|
29 |
{"role": "system", "content": "You are a helpful TCM medical assistant named 仲景中医大语言模型, created by 医哲未来 of Fudan University."},
|
30 |
{"role": "user", "content": prompt}
|
31 |
]
|
32 |
-
|
33 |
-
model_inputs = tokenizer([
|
34 |
generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=512)
|
35 |
generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
|
36 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
@@ -63,16 +63,13 @@ def multi_turn_chat(question, chat_history=None):
|
|
63 |
chat_history.append({"role": "assistant", "content": response})
|
64 |
|
65 |
# Format the chat history for output
|
66 |
-
tempass = ""
|
67 |
-
tempuser = ""
|
68 |
formatted_history = []
|
|
|
69 |
for entry in chat_history:
|
70 |
if entry['role'] == 'user':
|
71 |
tempuser = entry['content']
|
72 |
elif entry['role'] == 'assistant':
|
73 |
-
|
74 |
-
temp = (tempuser, tempass)
|
75 |
-
formatted_history.append(temp)
|
76 |
|
77 |
return formatted_history, chat_history
|
78 |
|
@@ -102,4 +99,4 @@ with gr.Blocks() as multi_turn_interface:
|
|
102 |
user_input.submit(multi_turn_chat, [user_input, state], [chatbot, state])
|
103 |
|
104 |
single_turn_interface.launch()
|
105 |
-
multi_turn_interface.launch()
|
|
|
12 |
# 初始化
|
13 |
peft_model_id = "CMLM/ZhongJing-2-1_8b"
|
14 |
base_model_id = "Qwen/Qwen1.5-1.8B-Chat"
|
15 |
+
model = AutoModelForCausalLM.from_pretrained(base_model_id, device_map="auto").to(device)
|
16 |
model.load_adapter(peft_model_id)
|
17 |
tokenizer = AutoTokenizer.from_pretrained(
|
18 |
"CMLM/ZhongJing-2-1_8b",
|
|
|
29 |
{"role": "system", "content": "You are a helpful TCM medical assistant named 仲景中医大语言模型, created by 医哲未来 of Fudan University."},
|
30 |
{"role": "user", "content": prompt}
|
31 |
]
|
32 |
+
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
33 |
+
model_inputs = tokenizer([input_text], return_tensors="pt").to(device)
|
34 |
generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=512)
|
35 |
generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
|
36 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
|
63 |
chat_history.append({"role": "assistant", "content": response})
|
64 |
|
65 |
# Format the chat history for output
|
|
|
|
|
66 |
formatted_history = []
|
67 |
+
tempuser = ""
|
68 |
for entry in chat_history:
|
69 |
if entry['role'] == 'user':
|
70 |
tempuser = entry['content']
|
71 |
elif entry['role'] == 'assistant':
|
72 |
+
formatted_history.append((tempuser, entry['content']))
|
|
|
|
|
73 |
|
74 |
return formatted_history, chat_history
|
75 |
|
|
|
99 |
user_input.submit(multi_turn_chat, [user_input, state], [chatbot, state])
|
100 |
|
101 |
single_turn_interface.launch()
|
102 |
+
multi_turn_interface.launch()
|