增加几个文件

exp_multimodal/text_embedder.py

@@ -0,0 +1,100 @@
+"""
+Ollama Embedding客户端
+调用Ollama的OpenAI兼容API获取文本向量表示
+"""
+import time
+from typing import List
+
+import numpy as np
+import requests
+
+
+class OllamaEmbedder:
+    """
+    基于Ollama部署的Embedding模型客户端
+    默认使用 quentinz/bge-large-zh-v1.5 (中文向量模型)
+    """
+    
+    def __init__(
+        self,
+        base_url: str = "http://192.168.1.250:11434",
+        model: str = "quentinz/bge-large-zh-v1.5",
+        timeout: int = 180,
+    ):
+        """
+        参数：
+            base_url: Ollama服务地址
+            model: Embedding模型名称
+            timeout: 请求超时时间（秒）
+        """
+        self.base_url = base_url.rstrip("/")
+        self.model = model
+        self.timeout = timeout
+        self.endpoint = f"{self.base_url}/v1/embeddings"
+        
+        print(f"[OllamaEmbedder] Initialized url={self.endpoint} model={model}")
+    
+    def encode(self, texts: List[str]) -> np.ndarray:
+        """
+        将文本列表编码为向量
+        
+        参数：
+            texts: 待编码的文本列表
+        
+        返回：
+            shape为(len(texts), embedding_dim)的numpy数组
+        """
+        if not texts:
+            return np.array([])
+        
+        t0 = time.time()
+        print(f"[OllamaEmbedder] Encoding {len(texts)} texts...")
+        
+        payload = {
+            "model": self.model,
+            "input": texts,
+        }
+        
+        try:
+            resp = requests.post(
+                self.endpoint,
+                json=payload,
+                timeout=self.timeout
+            )
+            dt = time.time() - t0
+            
+            if resp.status_code != 200:
+                print(f"[OllamaEmbedder] Error status={resp.status_code} body={resp.text[:200]}")
+                resp.raise_for_status()
+            
+            data = resp.json()
+            
+            # 解析OpenAI格式的响应: {"data": [{"embedding": [...]}, ...]}
+            if "data" not in data:
+                raise RuntimeError(f"Unexpected response format: {data}")
+            
+            embeddings = []
+            for item in data["data"]:
+                if "embedding" not in item:
+                    raise RuntimeError(f"Missing 'embedding' in response item: {item}")
+                embeddings.append(item["embedding"])
+            
+            result = np.array(embeddings, dtype=np.float32)
+            print(f"[OllamaEmbedder] Success shape={result.shape} elapsed={dt:.2f}s")
+            
+            return result
+            
+        except Exception as e:
+            dt = time.time() - t0
+            print(f"[OllamaEmbedder] Failed after {dt:.2f}s: {e}")
+            raise
+    
+    def encode_single(self, text: str) -> np.ndarray:
+        """
+        编码单个文本（便捷方法）
+        
+        返回：
+            shape为(embedding_dim,)的1D数组
+        """
+        result = self.encode([text])
+        return result[0] if len(result) > 0 else np.array([])