Feat/fix palyground bug (#655)

* fix playground bug, internet search judge

* fix playground internet bug

* modify delete mem

* modify tool resp bug in multi cube

* fix bug in playground chat handle and search inter

* modify prompt

* fix bug in playground

* fix bug playfround

* fix bug

* fix code

* fix model bug in playground

* modify plan b

* llm param modify

* add logger in playground

* modify code

* fix bug

* modify code

* modify code

* fix bug

---------

Co-authored-by: yuan.wang <[email protected]>
Co-authored-by: chunyu li <[email protected]>

Author: 3 people

src/memos/api/handlers/chat_handler.py

@@ -421,17 +421,21 @@ def generate_chat_response() -> Generator[str, None, None]:
                         query=chat_req.query,
                         user_id=chat_req.user_id,
                         readable_cube_ids=readable_cube_ids,
-                        mode=chat_req.mode,
+                        mode="fast",
                         internet_search=False,
-                        top_k=chat_req.top_k,
+                        top_k=5,
                         chat_history=chat_req.history,
                         session_id=chat_req.session_id,
-                        include_preference=chat_req.include_preference,
+                        include_preference=False,
                         pref_top_k=chat_req.pref_top_k,
                         filter=chat_req.filter,
+                        search_tool_memory=False,
                         playground_search_goal_parser=False,
                     )
+                    start_time = time.time()
                     search_response = self.search_handler.handle_search_memories(search_req)
+                    end_time = time.time()
+                    self.logger.info(f"first search time: {end_time - start_time}")
 
                     yield f"data: {json.dumps({'type': 'status', 'data': '1'})}\n\n"
 
@@ -447,18 +451,9 @@ def generate_chat_response() -> Generator[str, None, None]:
 
                     # Prepare reference data (first search)
                     reference = prepare_reference_data(filtered_memories)
-                    # get preference string
-                    pref_string = search_response.data.get("pref_string", "")
 
                     yield f"data: {json.dumps({'type': 'reference', 'data': reference})}\n\n"
 
-                    # Prepare preference markdown string
-                    if chat_req.include_preference:
-                        pref_list = search_response.data.get("pref_mem") or []
-                        pref_memories = pref_list[0].get("memories", []) if pref_list else []
-                        pref_md_string = self._build_pref_md_string_for_playground(pref_memories)
-                        yield f"data: {json.dumps({'type': 'pref_md_string', 'data': pref_md_string})}\n\n"
-
                     # parse goal for internet search
                     searcher = self.dependencies.searcher
                     parsed_goal = searcher.task_goal_parser.parse(
@@ -487,17 +482,22 @@ def generate_chat_response() -> Generator[str, None, None]:
                         or chat_req.query + (f"{parsed_goal.tags}" if parsed_goal.tags else ""),
                         user_id=chat_req.user_id,
                         readable_cube_ids=readable_cube_ids,
-                        mode=chat_req.mode,
+                        mode="fast",
                         internet_search=chat_req.internet_search,
                         top_k=chat_req.top_k,
                         chat_history=chat_req.history,
                         session_id=chat_req.session_id,
-                        include_preference=False,
+                        include_preference=chat_req.include_preference,
+                        pref_top_k=chat_req.pref_top_k,
                         filter=chat_req.filter,
                         search_memory_type="All",
+                        search_tool_memory=False,
                         playground_search_goal_parser=False,
                     )
+                    start_time = time.time()
                     search_response = self.search_handler.handle_search_memories(search_req)
+                    end_time = time.time()
+                    self.logger.info(f"second search time: {end_time - start_time}")
 
                     # Extract memories from search results (second search)
                     memories_list = []
@@ -516,12 +516,19 @@ def generate_chat_response() -> Generator[str, None, None]:
 
                     # Prepare remain reference data (second search)
                     reference = prepare_reference_data(filtered_memories)
+                    # get preference string
+                    pref_string = search_response.data.get("pref_string", "")
                     # get internet reference
                     internet_reference = self._get_internet_reference(
                         search_response.data.get("text_mem")[0]["memories"]
                     )
-
                     yield f"data: {json.dumps({'type': 'reference', 'data': reference})}\n\n"
+                    # Prepare preference markdown string
+                    if chat_req.include_preference:
+                        pref_list = search_response.data.get("pref_mem") or []
+                        pref_memories = pref_list[0].get("memories", []) if pref_list else []
+                        pref_md_string = self._build_pref_md_string_for_playground(pref_memories)
+                        yield f"data: {json.dumps({'type': 'pref_md_string', 'data': pref_md_string})}\n\n"
 
                     # Step 2: Build system prompt with memories
                     system_prompt = self._build_enhance_system_prompt(

src/memos/mem_reader/read_multi_modal/system_parser.py

@@ -1,5 +1,6 @@
 """Parser for system messages."""
 
+import ast
 import json
 import re
 import uuid
@@ -137,8 +138,14 @@ def parse_fine(
             tool_schema = json.loads(content)
             assert isinstance(tool_schema, list), "Tool schema must be a list[dict]"
         except json.JSONDecodeError:
-            logger.warning(f"[SystemParser] Failed to parse tool schema: {content}")
-            return []
+            try:
+                tool_schema = ast.literal_eval(content)
+                assert isinstance(tool_schema, list), "Tool schema must be a list[dict]"
+            except (ValueError, SyntaxError, AssertionError):
+                logger.warning(
+                    f"[SystemParser] Failed to parse tool schema with both JSON and ast.literal_eval: {content}"
+                )
+                return []
         except AssertionError:
             logger.warning(f"[SystemParser] Tool schema must be a list[dict]: {content}")
             return []

src/memos/memories/textual/tree.py

@@ -343,6 +343,17 @@ def delete_all(self) -> None:
             logger.error(f"An error occurred while deleting all memories: {e}")
             raise
 
+    def delete_by_filter(
+        self,
+        writable_cube_ids: list[str],
+        file_ids: list[str] | None = None,
+        filter: dict | None = None,
+    ) -> None:
+        """Delete memories by filter."""
+        self.graph_store.delete_node_by_prams(
+            writable_cube_ids=writable_cube_ids, file_ids=file_ids, filter=filter
+        )
+
     def load(self, dir: str) -> None:
         try:
             memory_file = os.path.join(dir, self.config.memory_filename)

src/memos/memories/textual/tree_text_memory/retrieve/searcher.py

@@ -701,15 +701,35 @@ def _sort_and_trim(
         """Sort results by score and trim to top_k"""
         final_items = []
         if search_tool_memory:
-            tool_results = [
+            tool_schema_results = [
                 (item, score)
                 for item, score in results
-                if item.metadata.memory_type in ["ToolSchemaMemory", "ToolTrajectoryMemory"]
+                if item.metadata.memory_type == "ToolSchemaMemory"
             ]
-            sorted_tool_results = sorted(tool_results, key=lambda pair: pair[1], reverse=True)[
-                :tool_mem_top_k
+            sorted_tool_schema_results = sorted(
+                tool_schema_results, key=lambda pair: pair[1], reverse=True
+            )[:tool_mem_top_k]
+            for item, score in sorted_tool_schema_results:
+                if plugin and round(score, 2) == 0.00:
+                    continue
+                meta_data = item.metadata.model_dump()
+                meta_data["relativity"] = score
+                final_items.append(
+                    TextualMemoryItem(
+                        id=item.id,
+                        memory=item.memory,
+                        metadata=SearchedTreeNodeTextualMemoryMetadata(**meta_data),
+                    )
+                )
+            tool_trajectory_results = [
+                (item, score)
+                for item, score in results
+                if item.metadata.memory_type == "ToolTrajectoryMemory"
             ]
-            for item, score in sorted_tool_results:
+            sorted_tool_trajectory_results = sorted(
+                tool_trajectory_results, key=lambda pair: pair[1], reverse=True
+            )[:tool_mem_top_k]
+            for item, score in sorted_tool_trajectory_results:
                 if plugin and round(score, 2) == 0.00:
                     continue
                 meta_data = item.metadata.model_dump()

src/memos/multi_mem_cube/single_cube.py

@@ -30,6 +30,7 @@
     SearchMode,
     UserContext,
 )
+from memos.utils import timed
 
 
 logger = get_logger(__name__)
@@ -198,6 +199,7 @@ def _get_search_mode(self, mode: str) -> str:
         """
         return mode
 
+    @timed
     def _search_text(
         self,
         search_req: APISearchRequest,
@@ -363,6 +365,7 @@ def _fine_search(
 
         return formatted_memories
 
+    @timed
     def _search_pref(
         self,
         search_req: APISearchRequest,
@@ -429,7 +432,7 @@ def _fast_search(
             top_k=search_req.top_k,
             mode=SearchMode.FAST,
             manual_close_internet=not search_req.internet_search,
-            momory_type=search_req.search_memory_type,
+            memory_type=search_req.search_memory_type,
             search_filter=search_filter,
             search_priority=search_priority,
             info={