gmakstutis/LightRAG
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat: enhance LLM output format tolerance for bracket processing

Browse source 
- Expand bracket tolerance to support additional characters: < > " '
- Implement symmetric handling for both leading and trailing characters
- Replace simple string matching with robust regex-based pattern detection
- Maintain full backward compatibility with existing bracket formats
This commit is contained in:
yangdx 2025-09-10 18:10:06 +08:00
parent 00de0a4be8
commit 02e7462645
2 changed files with 25 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

27
lightrag/operate.py
View file

@ -876,22 +876,39 @@ async def _process_extraction_result(
)
for record in records:
# Remove outer brackets (support English and Chinese brackets)
# Remove outer brackets (support English and Chinese brackets with enhanced tolerance)
record = record.strip()
# Define allowed leading and trailing characters
leading_trailing_chars = r'[`<>"\']*'
# Handle leading characters before left bracket
if record.startswith("(") or record.startswith("（"):
record = record[1:]
else:
if record.startswith("`(") or record.startswith("`（"):
record = record[2:]
# Check for leading characters + left bracket pattern
leading_bracket_pattern = r'^' + leading_trailing_chars + r'([（(])'
match = re.search(leading_bracket_pattern, record)
if match:
# Extract content from the left bracket position
bracket_pos = match.start(1)
record = record[bracket_pos + 1:]
else:
logger.warning(
f"{chunk_key}: Record starting bracket can not be found in extraction result"
)
# Handle trailing characters after right bracket
if record.endswith(")") or record.endswith("）"):
record = record[:-1]
else:
if record.endswith(")`") or record.endswith("）`"):
record = record[:-2]
# Check for right bracket + trailing characters pattern
trailing_bracket_pattern = r'([)）])' + leading_trailing_chars + r'$'
match = re.search(trailing_bracket_pattern, record)
if match:
# Extract content up to the right bracket position
bracket_pos = match.start(1)
record = record[:bracket_pos]
else:
logger.warning(
f"{chunk_key}: Record ending bracket can not be found in extraction result"

6
lightrag/prompt.py
View file

@ -28,8 +28,8 @@ You are a Knowledge Graph Specialist responsible for extracting entities and rel
5. **Relationship Order:** Prioritize relationships based on their significance to the intended meaning of input text, and output more crucial relationships first.
6. **Avoid Pronouns:** For entity names and all descriptions, explicitly name the subject or object instead of using pronouns; avoid pronouns such as `this document`, `our company`, `I`, `you`, and `he/she`.
7. **Undirectional Relationship:** Treat relationships as undirected; swapping the source and target entities does not constitute a new relationship. Avoid outputting duplicate relationships.
8. **Language:** Output entity names, keywords and descriptions in {language}.
9. **Delimiter:** Use `{record_delimiter}` as the entity or relationship list delimiter; output `{completion_delimiter}` when all the entities and relationships are extracted.
8. **Language:** Output entity names, keywords and descriptions in {language}. Proper nouns, such as personal names, should not be translated. Please keep them in their original language.
9. **Delimiter:** Use {record_delimiter} as the entity or relationship list delimiter; output {completion_delimiter} when all the entities and relationships are extracted.
---Examples---
{examples}
@ -49,7 +49,7 @@ Extract entities and relationships from the input text to be Processed.
---Instructions---
1. Output entities and relationships, prioritized by their relevance to the input text's core meaning.
2. Output `{completion_delimiter}` when all the entities and relationships are extracted.
3. Ensure the output language is {language}.
3. Ensure the output language is {language}. Proper nouns, such as personal names, should not be translated. Please keep them in their original language.
<Output>
"""