簡嘉琳
commited on
Commit
·
705c0ca
1
Parent(s):
5f37119
Fix trust_remote_code issue
Browse files
app.py
CHANGED
|
@@ -5,43 +5,45 @@ import torch
|
|
| 5 |
# 預先定義 Hugging Face 模型
|
| 6 |
MODEL_NAMES = {
|
| 7 |
"DeepSeek-V3": "deepseek-ai/DeepSeek-V3",
|
| 8 |
-
"DeepSeek-R1": "deepseek-ai/DeepSeek-R1"
|
| 9 |
}
|
| 10 |
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
|
|
|
| 16 |
return model, tokenizer
|
| 17 |
|
|
|
|
| 18 |
# 預設載入 DeepSeek-V3
|
| 19 |
current_model, current_tokenizer = load_model("DeepSeek-V3")
|
| 20 |
|
|
|
|
| 21 |
def chat(message, history, model_name):
|
| 22 |
"""處理聊天訊息"""
|
| 23 |
global current_model, current_tokenizer
|
| 24 |
-
|
| 25 |
# 若模型不同則切換
|
| 26 |
if model_name != current_model:
|
| 27 |
current_model, current_tokenizer = load_model(model_name)
|
| 28 |
-
|
| 29 |
inputs = current_tokenizer(message, return_tensors="pt").to("cuda")
|
| 30 |
outputs = current_model.generate(**inputs, max_length=1024)
|
| 31 |
response = current_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 32 |
-
|
| 33 |
return response
|
| 34 |
|
|
|
|
| 35 |
with gr.Blocks() as app:
|
| 36 |
gr.Markdown("## Chatbot with DeepSeek Models")
|
| 37 |
-
|
| 38 |
with gr.Row():
|
| 39 |
chat_interface = gr.ChatInterface(chat, streaming=True, save_history=True)
|
| 40 |
model_selector = gr.Dropdown(
|
| 41 |
-
choices=list(MODEL_NAMES.keys()),
|
| 42 |
-
value="DeepSeek-V3",
|
| 43 |
-
label="Select Model"
|
| 44 |
)
|
| 45 |
-
|
| 46 |
chat_interface.append(model_selector)
|
| 47 |
app.launch()
|
|
|
|
| 5 |
# 預先定義 Hugging Face 模型
|
| 6 |
MODEL_NAMES = {
|
| 7 |
"DeepSeek-V3": "deepseek-ai/DeepSeek-V3",
|
| 8 |
+
"DeepSeek-R1": "deepseek-ai/DeepSeek-R1",
|
| 9 |
}
|
| 10 |
|
| 11 |
+
|
| 12 |
+
def load_model(model_path):
|
| 13 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
| 14 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 15 |
+
model_path, torch_dtype=torch.float16, trust_remote_code=True
|
| 16 |
+
).cuda()
|
| 17 |
return model, tokenizer
|
| 18 |
|
| 19 |
+
|
| 20 |
# 預設載入 DeepSeek-V3
|
| 21 |
current_model, current_tokenizer = load_model("DeepSeek-V3")
|
| 22 |
|
| 23 |
+
|
| 24 |
def chat(message, history, model_name):
|
| 25 |
"""處理聊天訊息"""
|
| 26 |
global current_model, current_tokenizer
|
| 27 |
+
|
| 28 |
# 若模型不同則切換
|
| 29 |
if model_name != current_model:
|
| 30 |
current_model, current_tokenizer = load_model(model_name)
|
| 31 |
+
|
| 32 |
inputs = current_tokenizer(message, return_tensors="pt").to("cuda")
|
| 33 |
outputs = current_model.generate(**inputs, max_length=1024)
|
| 34 |
response = current_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 35 |
+
|
| 36 |
return response
|
| 37 |
|
| 38 |
+
|
| 39 |
with gr.Blocks() as app:
|
| 40 |
gr.Markdown("## Chatbot with DeepSeek Models")
|
| 41 |
+
|
| 42 |
with gr.Row():
|
| 43 |
chat_interface = gr.ChatInterface(chat, streaming=True, save_history=True)
|
| 44 |
model_selector = gr.Dropdown(
|
| 45 |
+
choices=list(MODEL_NAMES.keys()), value="DeepSeek-V3", label="Select Model"
|
|
|
|
|
|
|
| 46 |
)
|
| 47 |
+
|
| 48 |
chat_interface.append(model_selector)
|
| 49 |
app.launch()
|