Fix: table tag on chunks. (#12126)

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

agent/component/llm.py

@@ -56,7 +56,6 @@ class LLMParam(ComponentParamBase):
         self.check_nonnegative_number(int(self.max_tokens), "[Agent] Max tokens")
         self.check_decimal_float(float(self.top_p), "[Agent] Top P")
         self.check_empty(self.llm_id, "[Agent] LLM")
-        self.check_empty(self.sys_prompt, "[Agent] System prompt")
         self.check_empty(self.prompts, "[Agent] User prompt")
 
     def gen_conf(self):

api/db/services/document_service.py

@@ -696,10 +696,12 @@ class DocumentService(CommonService):
             for k,v in r.meta_fields.items():
                 if k not in meta:
                     meta[k] = {}
-                v = str(v)
-                if v not in meta[k]:
-                    meta[k][v] = []
-                meta[k][v].append(doc_id)
+                if not isinstance(v, list):
+                    v = [v]
+                for vv in v:
+                    if vv not in meta[k]:
+                        meta[k][vv] = []
+                    meta[k][vv].append(doc_id)
         return meta
 
     @classmethod

common/metadata_utils.py

@@ -44,21 +44,27 @@ def meta_filter(metas: dict, filters: list[dict], logic: str = "and"):
     def filter_out(v2docs, operator, value):
         ids = []
         for input, docids in v2docs.items():
+
             if operator in ["=", "≠", ">", "<", "≥", "≤"]:
                 try:
+                    if isinstance(input, list):
+                        input = input[0]
                     input = float(input)
                     value = float(value)
                 except Exception:
-                    input = str(input)
-                    value = str(value)
+                    pass
+            if isinstance(input, str):
+                input = input.lower()
+            if isinstance(value, str):
+                value = value.lower()
 
             for conds in [
-                (operator == "contains", str(value).lower() in str(input).lower()),
-                (operator == "not contains", str(value).lower() not in str(input).lower()),
-                (operator == "in", str(input).lower() in str(value).lower()),
-                (operator == "not in", str(input).lower() not in str(value).lower()),
-                (operator == "start with", str(input).lower().startswith(str(value).lower())),
-                (operator == "end with", str(input).lower().endswith(str(value).lower())),
+                (operator == "contains", input in value if not isinstance(input, list) else all([i in value for i in input])),
+                (operator == "not contains", input not in value if not isinstance(input, list) else all([i not in value for i in input])),
+                (operator == "in", input in value if not isinstance(input, list) else all([i in value for i in input])),
+                (operator == "not in", input not in value if not isinstance(input, list) else all([i not in value for i in input])),
+                (operator == "start with", str(input).lower().startswith(str(value).lower())  if not isinstance(input, list) else "".join([str(i).lower() for i in input]).startswith(str(value).lower())),
+                (operator == "end with", str(input).lower().endswith(str(value).lower())  if not isinstance(input, list) else "".join([str(i).lower() for i in input]).endswith(str(value).lower())),
                 (operator == "empty", not input),
                 (operator == "not empty", input),
                 (operator == "=", input == value),

rag/nlp/__init__.py

@@ -348,7 +348,8 @@ def tokenize_table(tbls, doc, eng, batch_size=10):
             d["doc_type_kwd"] = "table"
             if img:
                 d["image"] = img
-                d["doc_type_kwd"] = "image"
+                if d["content_with_weight"].find("<tr>") < 0:
+                    d["doc_type_kwd"] = "image"
             if poss:
                 add_positions(d, poss)
             res.append(d)
@@ -361,7 +362,8 @@ def tokenize_table(tbls, doc, eng, batch_size=10):
             d["doc_type_kwd"] = "table"
             if img:
                 d["image"] = img
-                d["doc_type_kwd"] = "image"
+                if d["content_with_weight"].find("<tr>") < 0:
+                    d["doc_type_kwd"] = "image"
             add_positions(d, poss)
             res.append(d)
     return res

rag/svr/task_executor.py

@@ -395,8 +395,8 @@ async def build_chunks(task, progress_callback):
             await asyncio.gather(*tasks, return_exceptions=True)
             raise
         metadata = {}
-        for ck in cks:
-            metadata = update_metadata_to(metadata, ck["metadata_obj"])
+        for doc in docs:
+            metadata = update_metadata_to(metadata, doc["metadata_obj"])
             del ck["metadata_obj"]
         if metadata:
             e, doc = DocumentService.get_by_id(task["doc_id"])