Update app.py

app.py

@@ -1,10 +1,54 @@
 import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModel
+import numpy as np
 
-with gr.Blocks(fill_height=True) as demo:
-    with gr.Sidebar():
-        gr.Markdown("# Inference Provider")
-        gr.Markdown("This Space showcases the anferico/bert-for-patents model, served by the hf-inference API. Sign in with your Hugging Face account to use this API.")
-        button = gr.LoginButton("Sign in")
-    gr.load("models/anferico/bert-for-patents", accept_token=button, provider="hf-inference")
+# 1. 加载谷歌专利专用 BERT (0.3B 极其轻量)
+model_name = "anferico/bert-for-patents"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModel.from_pretrained(model_name)
+
+def get_embedding(text):
+    # 将文本转化为 1024 维的专利语义向量
+    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
+    with torch.no_grad():
+        outputs = model(**inputs)
+    # 取最后一层隐藏状态的平均值作为特征向量 (Mean Pooling)
+    embeddings = outputs.last_hidden_state.mean(dim=1).squeeze().numpy()
+    return embeddings.tolist()
+
+def compute_similarity(text1, text2):
+    # 计算两个专利描述之间的语义相似度 (余弦相似度)
+    v1 = np.array(get_embedding(text1))
+    v2 = np.array(get_embedding(text2))
+    similarity = np.dot(v1, v2) / (np.linalg.norm(v1) * np.linalg.norm(v2))
+    return f"{similarity:.4f}"
+
+# 2. 界面装修：走 OpenIPOS 旗舰风格
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.HTML("""
+        <div style="display: flex; align-items: center; gap: 20px; padding: 10px;">
+            <img src="https://s3.bmp.ovh/2026/03/30/rIAGKg0O.png" style="height: 50px;">
+            <div>
+                <h1 style="margin:0;">OpenIPOS Patent Vector Engine</h1>
+                <p style="margin:0; color:#666;">基于 BERT-for-Patents 的专业专利语义标尺 (0.3B)</p>
+            </div>
+        </div>
+    """)
     
+    with gr.Tab("语义相似度比对"):
+        t1 = gr.Textbox(label="技术特征 A (或权利要求 1)", placeholder="例如：一种含有纳米材料的涂层...")
+        t2 = gr.Textbox(label="技术特征 B (或对比文件 D1)", placeholder="例如：一种表面处理工艺，包含特殊涂层...")
+        btn = gr.Button("开始比对", variant="primary")
+        out = gr.Label(label="语义重合度 (0-1)")
+        btn.click(compute_similarity, inputs=[t1, t2], outputs=out)
+        
+    with gr.Tab("向量化测试 (API 用)"):
+        input_text = gr.Textbox(label="输入专利文本")
+        vector_out = gr.JSON(label="生成的 1024 维特征向量")
+        v_btn = gr.Button("生成向量")
+        v_btn.click(get_embedding, inputs=input_text, outputs=vector_out)
+
+    gr.Markdown("> **提示**：该 Space 已自动开启 API 访问。你可以将此 URL 对接到 Dify 的自定义算子中，作为专利检索的 Embedding 引擎。")
+
 demo.launch()