Spaces:
Sleeping
Sleeping
File size: 7,185 Bytes
c6868fa | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 | from pathlib import Path
import json
from typing import List
from langchain_core.documents import Document
from langchain_text_splitters import (
RecursiveCharacterTextSplitter,
MarkdownHeaderTextSplitter,
RecursiveJsonSplitter,
)
from chonkie import CodeChunker
from config import CHUNK_OVERLAP,CHUNK_SIZE,AST_BASED_SPLITTING
def custom_splitter(docs: List[Document],current_dir: Path) -> List[Document]:
all_chunks: List[Document] = []
md_splitter = MarkdownHeaderTextSplitter(
headers_to_split_on=[("#", "H1"), ("##", "H2"), ("###", "H3")]
)
text_fallback_splitter = RecursiveCharacterTextSplitter(
chunk_size=CHUNK_SIZE,
chunk_overlap=CHUNK_OVERLAP,
)
json_splitter = RecursiveJsonSplitter(
max_chunk_size=CHUNK_SIZE,
)
csv_splitter = RecursiveCharacterTextSplitter(
separators=["\n"],
chunk_size=CHUNK_SIZE,
chunk_overlap=0,
)
for doc in docs:
# --- FIX: Empty Files Check ---
# Skip completely empty documents to save compute time
if not doc.page_content or not doc.page_content.strip():
continue
source_str = doc.metadata.get("source", "")
if not source_str:
continue
path = Path(source_str)
ext = path.suffix.lower()
try:
repo_path = str(path.relative_to(current_dir))
except ValueError:
repo_path = str(path)
base_metadata = {
**doc.metadata,
"file_name": path.name,
"extension": ext,
"path_rel_repo": repo_path,
}
doc_chunks: List[Document] = []
# AST-based code chunking
if ext in AST_BASED_SPLITTING:
ast_chunker = CodeChunker(
language=AST_BASED_SPLITTING.get(ext),
tokenizer="character",
chunk_size=CHUNK_SIZE,
include_nodes=False,
)
try:
chonkie_chunks = ast_chunker.chunk(doc.page_content)
for chunk in chonkie_chunks:
doc_chunks.append(
Document(
page_content=chunk.text,
metadata=base_metadata.copy(),
)
)
except Exception as e:
print(
f"Warning: AST parsing failed for {path.name}. "
f"Falling back to text. Error: {e}"
)
doc_chunks = text_fallback_splitter.split_documents([doc])
# Markdown
elif ext in {".md", ".mdx"}:
md_splits = md_splitter.split_text(doc.page_content)
for split in md_splits:
split.metadata = {**base_metadata, **split.metadata}
doc_chunks = text_fallback_splitter.split_documents(md_splits)
# JSON
elif ext == ".json":
try:
parsed_data = json.loads(doc.page_content)
#------ Normalize the data: because remeber json can be in two formate one single dictionary or list of dictionary
texts_to_split = []
if isinstance(parsed_data, list):
# If it's a list, treat each item as a separate document
# This yields much better search results for RAG
for item in parsed_data:
if isinstance(item, dict):
texts_to_split.append(item)
else:
texts_to_split.append({"value": item})
elif isinstance(parsed_data, dict):
# If it's already a dict, it's safe
texts_to_split.append(parsed_data)
else:
# If it's a raw string/number/bool
texts_to_split.append({"value": parsed_data})
# ---------------------------------------------
# Create metadatas array to match the length of texts_to_split
metadatas = [base_metadata.copy() for _ in texts_to_split]
json_docs = json_splitter.create_documents(
texts=texts_to_split,
metadatas=metadatas,
)
doc_chunks.extend(json_docs)
except json.JSONDecodeError as e:
print(
f"Warning: Invalid JSON syntax in {path.name}. "
f"Falling back to text. Error: {e}"
)
doc_chunks = text_fallback_splitter.split_documents([doc])
# JSONL
elif ext == ".jsonl":
for line in doc.page_content.splitlines():
line = line.strip()
if not line:
continue
try:
line_data = json.loads(line)
# --- Normalize JSONL lines ---
if not isinstance(line_data, dict):
line_data = {"value": line_data}
json_docs = json_splitter.create_documents(
texts=[line_data],
metadatas=[base_metadata.copy()],
)
doc_chunks.extend(json_docs)
except json.JSONDecodeError as e:
print(
f"Warning: Invalid JSONL line in {path.name}. "
f"Skipping. Error: {e}"
)
# CSV / TSV
elif ext in {".csv", ".tsv"}:
lines = doc.page_content.splitlines()
if not lines:
continue
header = lines[0]
doc_chunks = csv_splitter.split_documents([doc])
for i, chunk in enumerate(doc_chunks):
if i == 0:
continue
# --- FIX: CSV Header Logic ---
# Ensure the chunk doesn't already have the header and strip leading newlines
# to prevent broken/malformed line boundaries.
if not chunk.page_content.startswith(header):
chunk.page_content = header + "\n" + chunk.page_content.lstrip()
chunk.metadata = base_metadata.copy()
# Fallback
else:
doc_chunks = text_fallback_splitter.split_documents([doc])
# ── FILE NAME INJECTION ───────────────────────────────────────────────
# Inject the file name into the text payload to give LLM Context.
for chunk in doc_chunks:
# 1. Update metadata
chunk.metadata = {**base_metadata, **chunk.metadata}
chunk.page_content = f"[FILE: {path.name}]\n\n" + chunk.page_content
all_chunks.append(chunk)
print(f"Original Files Processed : {len(docs)}")
print(f"Total Chunks Generated : {len(all_chunks)}")
return all_chunks
|