base90-rag 0.2.2__tar.gz → 0.2.3__tar.gz

{base90_rag-0.2.2 → base90_rag-0.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: base90-rag
-Version: 0.2.2
+Version: 0.2.3
 Summary: 超极速、零算力、零幻觉的纯文本 RAG 检索结构 (Base90 极简硬路由)
 Author-email: Flowerdance <admin@example.com>
 License: MIT

{base90_rag-0.2.2 → base90_rag-0.2.3}/base90_rag/core.py

@@ -204,47 +204,58 @@ class Base90Memory:
         valid_sets = []
         for code in query_codes:
             if code in self.mem_inverted:
-                word = self.code_to_word.get(code, "")
-                if len(word) == 1 and word in "的了是我你他她它": continue
-                if len(word) == 1 and word.isascii() and word.isalpha(): continue
-
-                valid_sets.append(self.mem_inverted[code])
+                inv_set = self.mem_inverted[code]
+                if len(inv_set) > 5000:
+                    from itertools import islice
+                    inv_set = set(islice(inv_set, 5000))
+                valid_sets.append(inv_set)
 
         if not valid_sets: return []
 
         valid_sets.sort(key=len)
-
         candidate_pids = valid_sets[0].copy()
         for i in range(1, len(valid_sets)):
             candidate_pids = candidate_pids.intersection(valid_sets[i])
             if not candidate_pids:
                 break
 
-        if len(candidate_pids) > 2000:
-            from itertools import islice
-            candidate_pids = set(islice(candidate_pids, 2000))
+        import re
+        clean_query = re.sub(r'\s+', '', query).lower()
 
         final_results = []
 
+        def _is_match(raw_text):
+            if query in raw_text:
+                return True
+
+            if clean_query in raw_text.replace(" ", "").lower():
+                return True
+            return False
+
         if candidate_pids:
+            check_count = 0
             for pid in candidate_pids:
                 v = self.mem_vaults.get(pid)
-                if v and query in v["text"]:
+                if v and _is_match(v["text"]):
                     final_results.append(v["text"])
                     if len(final_results) >= limit:
                         return final_results
+                check_count += 1
+                if check_count > 2000:
+                    break
 
-        count = 0
-        for pid in valid_sets[0]:
-            v = self.mem_vaults.get(pid)
-            if v and v["text"] not in final_results:  # 去重
-                if query in v["text"]:  # 硬校验
-                    final_results.append(v["text"])
-                    if len(final_results) >= limit:
-                        break
-            count += 1
-            if count > 2000:
-                break
+        if len(final_results) < limit:
+            check_count = 0
+            for pid in valid_sets[0]:
+                v = self.mem_vaults.get(pid)
+                if v and v["text"] not in final_results:
+                    if _is_match(v["text"]):
+                        final_results.append(v["text"])
+                        if len(final_results) >= limit:
+                            break
+                check_count += 1
+                if check_count > 300:
+                    break
 
         return final_results
 

{base90_rag-0.2.2 → base90_rag-0.2.3}/base90_rag.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: base90-rag
-Version: 0.2.2
+Version: 0.2.3
 Summary: 超极速、零算力、零幻觉的纯文本 RAG 检索结构 (Base90 极简硬路由)
 Author-email: Flowerdance <admin@example.com>
 License: MIT

{base90_rag-0.2.2 → base90_rag-0.2.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "base90-rag"
-version = "0.2.2"
+version = "0.2.3"
 authors = [
   { name="Flowerdance", email="admin@example.com" }
 ]