Spaces:
Sleeping
Sleeping
File size: 7,508 Bytes
4225666 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 | from langchain_core.documents import Document
from pathlib import Path
from typing import Optional, List
from datetime import datetime, date
import uuid
import yaml
from app.services.text_splitter import TextSplitter
import json
# Allowed types for metadata cleaning
ALLOWED = (str, int, float, bool, list, type(None))
def get_references_v2(docs, threshold: float):
results = []
context = ""
for doc in docs:
_doc = doc.document
_similarity = doc.fused_score
# print(_similarity, threshold)
if _similarity < threshold:
continue
metadata = _doc.metadata
document = {
"title": metadata.get("title", metadata.get("name", metadata.get("topic", "untitled"))),
"chunk_index": metadata.get("chunk_index"),
"source": metadata.get("source_file", metadata.get("source", "untitled")),
"page_content": _doc.page_content,
"similarity": _similarity
}
ctx = f"""{document['title']} page_content: {document['page_content']}, from source: {document['source']}.\n\n"""
context += ctx
results.append(document)
return {
"documents": results,
"context": context
}
def get_references(docs, threshold: float):
results = []
context = ""
for doc in docs:
_doc = doc[0]
_similarity = 1 - doc[1]
if _similarity < threshold:
continue
metadata = _doc.metadata
document = {
"title": metadata.get("title", metadata.get("name", metadata.get("topic", "untitled"))),
"chunk_index": metadata.get("chunk_index"),
"source": metadata.get("source_file", metadata.get("source", "untitled")),
"page_content": _doc.page_content,
"similarity": _similarity
}
ctx = f"""
page_content: {document['page_content']}, from source: {document['source']}.
"""
context += ctx
results.append(document)
return {
"documents": results,
"context": context
}
def create_documents(
chunks: List[str],
filePath: Optional[Path] = None,
built_in_metadata: Optional[dict] = {},
title: Optional[str] = None
) -> List[Document]:
"""
Create Document objects from text chunks with standard metadata (UUIDs, timestamps, indices).
Works for both files (filePath provided) and raw text (filePath=None).
"""
if filePath and filePath.exists():
created_date = datetime.fromtimestamp(filePath.stat().st_ctime).isoformat()
modified_date = datetime.fromtimestamp(filePath.stat().st_mtime).isoformat()
source = filePath.name
given_title = title or filePath.stem
else:
now = datetime.now().isoformat()
created_date = now
modified_date = now
# Use existing source from metadata if available, else empty
source = built_in_metadata.get("source", "")
if not source and filePath:
source = filePath.name
given_title = title or built_in_metadata.get("title", "Untitled")
docs = []
for i, chunk in enumerate(chunks):
# Base metadata
metadata = {
"doc_id": str(uuid.uuid4()), # unique chunk id
"source": source,
"title": given_title,
"created_date": created_date,
"modified_date": modified_date,
"chunk_index": i,
}
# Merge built-in, but don't overwrite our system fields if they exist
# actually, built-in should probably take precedence for some things?
# Let's simple merge:
metadata.update(built_in_metadata)
# Ensure our critical fields are set correctly after merge (if built-in had conflict)
metadata["doc_id"] = metadata.get("doc_id", str(uuid.uuid4()))
metadata["chunk_index"] = i
doc = Document(page_content=chunk, metadata=metadata)
docs.append(doc)
return docs
def create_document(
text: str,
metadata: dict
):
return Document(page_content=text, metadata=metadata)
def clean_metadata(metadata: dict):
cleaned = {}
for k, v in metadata.items():
if isinstance(v, (datetime, date)):
cleaned[k] = v.isoformat()
elif isinstance(v, ALLOWED):
cleaned[k] = v
else:
cleaned[k] = str(v)
return cleaned
def read_text_file(filePath: Path):
with open(filePath, "r", encoding="utf-8") as f:
content = f.read()
return content
def read_json_file(filePath: Path):
with open(filePath, 'r') as file:
data = json.load(file)
return data
def build_metadata(filePath: Optional[Path] = None, content: Optional[str] = None):
if filePath:
content = read_text_file(filePath)
parts = content.split("---", 2)
if len(parts) >= 3:
frontmatter = yaml.safe_load(parts[1]) or {}
frontmatter = clean_metadata(frontmatter)
# add file name as source always
if filePath:
frontmatter["source"] = filePath.name
elif "source" not in frontmatter:
frontmatter["source"] = ""
return {
"metadata": frontmatter,
"content": parts[2].strip()
}
else:
# Don't enforce empty source if not provided, allows external metadata to stick
meta = {}
if filePath:
meta["source"] = filePath.name
return {
"metadata": meta,
"content": content.strip()
}
def create_documents_from_text(text: str, metadata: dict = {}):
"""
Create documents from raw text with automatic splitting and metadata enrichment.
"""
text = text.strip()
data = build_metadata(content=text)
# 1. Smart Metadata Merge
final_metadata = data["metadata"].copy()
# Update with provided metadata
if final_metadata.get("source") == "" and metadata.get("source"):
final_metadata["source"] = metadata["source"]
# Merge regular keys
final_metadata.update({k:v for k,v in metadata.items() if k != "source"})
text = data["content"]
# 2. Split text into chunks (strings)
# Use section-aware splitter if text contains markdown section delimiters
if "\n---\n" in text or text.startswith("---\n"):
splitter = TextSplitter.for_markdown_with_sections()
else:
splitter = TextSplitter()
chunks = splitter.split_text(text)
# 3. Create documents using standard helper (adds IDs, indices, dates)
return create_documents(
chunks=chunks,
filePath=None,
built_in_metadata=final_metadata
)
def load_json(filePath: Path):
data = read_json_file(filePath=filePath)
filePath = Path(filePath)
file_name = filePath.name
metadata = {
"id": data["id"],
"title": data.get("name", data.get("title", "Untitled")),
"source": data["source"],
"source_file": file_name or "Untitled",
"created_date": datetime.now().isoformat()
}
docs= []
splitter = TextSplitter()
for key,value in data["content"].items():
ctx = splitter.split_text(value.strip())
for idx, chunk in enumerate(ctx):
if(chunk.strip() == ""):
continue
else:
chunk = f"{key}: {chunk.strip()}"
docs.append(Document(page_content=chunk, metadata={**metadata, "topic": key, "chunk_index": idx}))
return docs
|