.

app/core/embedding.py

@@ -27,5 +27,9 @@ class BGEEmbedding:
         input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
         return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
 
+    def get_agg_embedding_from_queries(self, queries):
+        embeddings = self.encode(queries)
+        return np.mean(embeddings, axis=0)  # 聚合多个嵌入
+    
 embedder = BGEEmbedding()