Spaces:
Running on Zero
Running on Zero
zhang commited on
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,57 +3,44 @@ import gradio as gr
|
|
| 3 |
import spaces
|
| 4 |
from huggingface_hub import InferenceClient
|
| 5 |
|
| 6 |
-
# 1.
|
| 7 |
hf_token = os.getenv("HF_TOKEN")
|
| 8 |
-
|
| 9 |
-
# 2. 初始化客户端,指向 DeepSeek-R1-32B 满血版
|
| 10 |
-
# 如果没有设置 HF_TOKEN,这里会报错,所以我们在下面加了异常处理
|
| 11 |
client = InferenceClient("deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", token=hf_token)
|
| 12 |
|
| 13 |
-
# 3. 定义核心推理逻辑
|
| 14 |
@spaces.GPU
|
| 15 |
def chat_fn(message, history):
|
| 16 |
-
# 构建对话历史上下文
|
| 17 |
messages = []
|
| 18 |
for turn in history:
|
| 19 |
-
# 兼容不同版本的 Gradio 历史记录格式
|
| 20 |
if isinstance(turn, dict):
|
| 21 |
messages.append({"role": turn["role"], "content": turn["content"]})
|
| 22 |
else:
|
| 23 |
messages.append({"role": "user", "content": turn[0]})
|
| 24 |
messages.append({"role": "assistant", "content": turn[1]})
|
| 25 |
-
|
| 26 |
messages.append({"role": "user", "content": message})
|
| 27 |
|
| 28 |
response = ""
|
| 29 |
-
# 开启流式输出 (Stream),让字一个一个跳出来
|
| 30 |
try:
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
yield response
|
| 40 |
except Exception as e:
|
| 41 |
-
yield f"⚠️ OpenIPOS
|
| 42 |
|
| 43 |
-
#
|
| 44 |
with gr.Blocks(fill_height=True, theme=gr.themes.Soft()) as demo:
|
| 45 |
gr.Markdown("# 🚀 OpenIPOS Global 算力节点")
|
| 46 |
gr.Markdown("### ⚖️ DeepSeek-R1 32B 法律级满血演示站")
|
| 47 |
-
gr.Markdown("本节点由 **OpenIPOS** 部署于全球高算力数据中心,支持完整 R1 思考链路,专为专利分析与法律逻辑设计。")
|
| 48 |
|
| 49 |
-
# 核心聊天界面
|
| 50 |
gr.ChatInterface(
|
| 51 |
chat_fn,
|
| 52 |
-
examples=["请分析这篇专利要求书的逻辑漏洞...", "对比香港与内地的发明专利申请流程"
|
| 53 |
)
|
| 54 |
-
|
| 55 |
gr.Markdown("---")
|
| 56 |
-
gr.Markdown("© 2026 OpenIPOS Global Limited.
|
| 57 |
|
| 58 |
-
# 5. 启动程序
|
| 59 |
demo.launch()
|
|
|
|
| 3 |
import spaces
|
| 4 |
from huggingface_hub import InferenceClient
|
| 5 |
|
| 6 |
+
# 1. 读取密钥
|
| 7 |
hf_token = os.getenv("HF_TOKEN")
|
|
|
|
|
|
|
|
|
|
| 8 |
client = InferenceClient("deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", token=hf_token)
|
| 9 |
|
|
|
|
| 10 |
@spaces.GPU
|
| 11 |
def chat_fn(message, history):
|
|
|
|
| 12 |
messages = []
|
| 13 |
for turn in history:
|
|
|
|
| 14 |
if isinstance(turn, dict):
|
| 15 |
messages.append({"role": turn["role"], "content": turn["content"]})
|
| 16 |
else:
|
| 17 |
messages.append({"role": "user", "content": turn[0]})
|
| 18 |
messages.append({"role": "assistant", "content": turn[1]})
|
|
|
|
| 19 |
messages.append({"role": "user", "content": message})
|
| 20 |
|
| 21 |
response = ""
|
|
|
|
| 22 |
try:
|
| 23 |
+
# 这里是核心优化:增加对异常块的过滤
|
| 24 |
+
for msg in client.chat_completion(messages, max_tokens=4096, stream=True):
|
| 25 |
+
# 只有当返回结构完整且有内容时才处理
|
| 26 |
+
if len(msg.choices) > 0:
|
| 27 |
+
token = msg.choices[0].delta.content
|
| 28 |
+
if token:
|
| 29 |
+
response += token
|
| 30 |
+
yield response
|
|
|
|
| 31 |
except Exception as e:
|
| 32 |
+
yield f"⚠️ OpenIPOS 调度提醒: {str(e)}"
|
| 33 |
|
| 34 |
+
# UI 部分保持你的 OpenIPOS 品牌风格
|
| 35 |
with gr.Blocks(fill_height=True, theme=gr.themes.Soft()) as demo:
|
| 36 |
gr.Markdown("# 🚀 OpenIPOS Global 算力节点")
|
| 37 |
gr.Markdown("### ⚖️ DeepSeek-R1 32B 法律级满血演示站")
|
|
|
|
| 38 |
|
|
|
|
| 39 |
gr.ChatInterface(
|
| 40 |
chat_fn,
|
| 41 |
+
examples=["请分析这篇专利要求书的逻辑漏洞...", "对比香港与内地的发明专利申请流程"],
|
| 42 |
)
|
|
|
|
| 43 |
gr.Markdown("---")
|
| 44 |
+
gr.Markdown("© 2026 OpenIPOS Global Limited. 算力驱动:Hugging Face ZeroGPU")
|
| 45 |
|
|
|
|
| 46 |
demo.launch()
|