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

Simplify tuple delimiter regex patterns for LLM output fixing

Browse source 
• Consolidate 6 regex patterns into 3
• More efficient pattern matching
• Clearer comments and examples
• Same functionality, less code
• Better maintainability
This commit is contained in:
yangdx 2025-09-12 00:56:40 +08:00
parent 78eadc1d6c
commit b9f80263b8
1 changed files with 12 additions and 39 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

51
lightrag/operate.py
View file

@ -877,63 +877,36 @@ async def _process_extraction_result(
# Fix various forms of tuple_delimiter corruption from the LLM output. # Fix various forms of tuple_delimiter corruption from the LLM output.
# It handles missing or replaced characters around the core delimiter. # It handles missing or replaced characters around the core delimiter.
# 1. `<` or `>` may be missing. # 1. There might be extra characters inserted between the bracket and pipeline.
# 2. `|` may be missing or replaced by another character. # 2. `|` may be missing or replaced by another character.
# 3. There might be extra characters inserted. # 3. Missing opening `<` or closing `>`
# 4. Missing opening `<` or closing `>`
# Example transformations: # Example transformations:
# <SEP> -> <|SEP|> # <X|SEP|> -> <|SEP|>, <|SEP|Y> -> <|SEP|>, <X|SEP|Y> -> <|SEP|> ((one extra characters outside pipes)
# <SEP|> -> <|SEP|> (where left | is missing) # <SEP>, <SEP|>, <|SEP> -> <|SEP|> (missing one or both pipes)
# <|SEP> -> <|SEP|> (where right | is missing) # <XSEP|> -> <|SEP|>, <|SEPX> -> <|SEP|> (where one | is replace by other charater)
# <XSEP|> -> <|SEP|> (where left | is replace by other charater) # |SEP|> -> <|SEP|>, <|SEP| -> <|SEP|> (where one | is missing)
# <|SEPX> -> <|SEP|> (where right | is replace by other charater)
# <|SEP|X> -> <|SEP|> (where X is not '>')
# <XX|SEP|YY> -> <|SEP|> (handles extra characters)
# |SEP|> -> <|SEP|> (where left | is missing)
# <|SEP| -> <|SEP|> (where right | is missing)
escaped_delimiter_core = re.escape( escaped_delimiter_core = re.escape(
tuple_delimiter[2:-2] tuple_delimiter[2:-2]
) # Extract "SEP" from "<|SEP|>" ) # Extract "SEP" from "<|SEP|>"
# Fix: <SEP> -> <|SEP|> (missing pipes) # Fix: <X|SEP|> -> <|SEP|>, <|SEP|Y> -> <|SEP|>, <X|SEP|Y> -> <|SEP|> (one extra characters outside pipes)
record = re.sub( record = re.sub(
rf"<{escaped_delimiter_core}>", rf"<.?\|{escaped_delimiter_core}\|.?>",
tuple_delimiter, tuple_delimiter,
record, record,
) )
# Fix: <SEP|> -> <|SEP|> (missing left pipe only) # Fix: <SEP>, <SEP|>, <|SEP> -> <|SEP|> (missing one or both pipes)
record = re.sub( record = re.sub(
rf"<{escaped_delimiter_core}\|>", rf"<\|?{escaped_delimiter_core}\|?>",
tuple_delimiter, tuple_delimiter,
record, record,
) )
# Fix: <|SEP> -> <|SEP|> (missing right pipe only) # Fix: <XSEP|> -> <|SEP|>, <|SEPX> -> <|SEP|> (one pipe is replaced by other character)
record = re.sub( record = re.sub(
rf"<\|{escaped_delimiter_core}>", rf"<[^|]{escaped_delimiter_core}\|>|<\|{escaped_delimiter_core}[^|]>",
tuple_delimiter,
record,
)
# Fix: <XSEP|> -> <|SEP|> (character X replacing first pipe)
record = re.sub(
rf"<[^|]+{escaped_delimiter_core}\|>",
tuple_delimiter,
record,
)
# Fix: <|SEPX> -> <|SEP|> (character X replacing second pipe)
record = re.sub(
rf"<\|{escaped_delimiter_core}[^|]+>",
tuple_delimiter,
record,
)
# Fix: <XX|SEP|YY> -> <|SEP|> (extra characters around, but preserve correct delimiters)
record = re.sub(
rf"<[^<>]+\|{escaped_delimiter_core}\|[^<>]+>",
tuple_delimiter, tuple_delimiter,
record, record,
) )