Spaces:
Running on Zero
Running on Zero
zhang commited on
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,16 +1,22 @@
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import spaces
|
| 3 |
from huggingface_hub import InferenceClient
|
| 4 |
|
| 5 |
-
#
|
| 6 |
-
|
| 7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
@spaces.GPU
|
| 9 |
def chat_fn(message, history):
|
| 10 |
-
# 构建对话
|
| 11 |
messages = []
|
| 12 |
for turn in history:
|
| 13 |
-
#
|
| 14 |
if isinstance(turn, dict):
|
| 15 |
messages.append({"role": turn["role"], "content": turn["content"]})
|
| 16 |
else:
|
|
@@ -20,11 +26,11 @@ def chat_fn(message, history):
|
|
| 20 |
messages.append({"role": "user", "content": message})
|
| 21 |
|
| 22 |
response = ""
|
| 23 |
-
#
|
| 24 |
try:
|
| 25 |
for msg in client.chat_completion(
|
| 26 |
messages,
|
| 27 |
-
max_tokens=
|
| 28 |
stream=True,
|
| 29 |
):
|
| 30 |
token = msg.choices[0].delta.content
|
|
@@ -32,17 +38,22 @@ def chat_fn(message, history):
|
|
| 32 |
response += token
|
| 33 |
yield response
|
| 34 |
except Exception as e:
|
| 35 |
-
yield f"
|
| 36 |
|
| 37 |
-
# 构建 OpenIPOS 品牌界面
|
| 38 |
-
with gr.Blocks(fill_height=True) as demo:
|
| 39 |
gr.Markdown("# 🚀 OpenIPOS Global 算力节点")
|
| 40 |
-
gr.Markdown("### DeepSeek-R1 32B 满血
|
|
|
|
| 41 |
|
| 42 |
-
#
|
| 43 |
gr.ChatInterface(
|
| 44 |
chat_fn,
|
| 45 |
-
|
| 46 |
)
|
| 47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
demo.launch()
|
|
|
|
| 1 |
+
import os
|
| 2 |
import gradio as gr
|
| 3 |
import spaces
|
| 4 |
from huggingface_hub import InferenceClient
|
| 5 |
|
| 6 |
+
# 1. 自动读取你在 Settings 里面设置的保险箱密钥
|
| 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:
|
|
|
|
| 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 |
for msg in client.chat_completion(
|
| 32 |
messages,
|
| 33 |
+
max_tokens=4096, # 调大 token,支持长文本专利分析
|
| 34 |
stream=True,
|
| 35 |
):
|
| 36 |
token = msg.choices[0].delta.content
|
|
|
|
| 38 |
response += token
|
| 39 |
yield response
|
| 40 |
except Exception as e:
|
| 41 |
+
yield f"⚠️ OpenIPOS 算力调度异常: {str(e)}。请确保已在 Settings 中设置 HF_TOKEN。"
|
| 42 |
|
| 43 |
+
# 4. 构建 OpenIPOS 品牌演示界面
|
| 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", "帮我写一段专利侵权判定的法律意见书"],
|
| 53 |
)
|
| 54 |
|
| 55 |
+
gr.Markdown("---")
|
| 56 |
+
gr.Markdown("© 2026 OpenIPOS Global Limited. All Rights Reserved. 算力驱动:Hugging Face ZeroGPU")
|
| 57 |
+
|
| 58 |
+
# 5. 启动程序
|
| 59 |
demo.launch()
|