Spaces:
Sleeping
Sleeping
File size: 14,737 Bytes
16d5a75 744b763 16d5a75 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 |
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_core.documents import Document
from langchain_core.output_parsers import StrOutputParser
from PIL import Image
import io
import fitz # PyMuPDF
from docx import Document as DocxDoc
import os
import uuid
import tempfile
import math
import base64
from src.config.cloudinary import upload_image
from src.data_preprocessing.prompt import image_caption_prompt
from src.config.llm import llm_2_0 as llm
from src.utils.logger import logger
def extract_and_chunk_documents(
file_path: str,
chunk_size: int = 1000,
chunk_overlap: int = 200,
upload_images: bool = True,
batch_size: int = 15,
):
"""
1. Extract text and images from the document, keeping them in the order they appear.
2. Upload images to Cloudinary and get captions using the image_caption_prompt in batches.
3. Create separate chunks for text and images.
Args:
file_path: Path to the document file
chunk_size: Size of text chunks
chunk_overlap: Overlap between chunks
upload_images: Whether to upload images to Cloudinary
batch_size: Number of images to process in a single batch
Returns:
List of Document objects with separate text and image chunks
"""
docs = [] # Store documents (both text and images)
image_caption_chain = image_caption_prompt | llm | StrOutputParser()
# Extract text and images from document
if file_path.endswith(".docx"):
docs = extract_docx_with_images(file_path)
elif file_path.endswith(".pdf"):
docs = extract_pdf_with_images(file_path)
else:
raise ValueError("Unsupported file type")
# Separate text and image documents
text_docs = [doc for doc in docs if doc.metadata.get("type") == "text"]
image_docs = [doc for doc in docs if doc.metadata.get("type") == "image"]
# Process images in batches: upload to Cloudinary and get captions
processed_image_chunks = []
if upload_images and image_docs:
# Prepare image batches
total_images = len(image_docs)
num_batches = math.ceil(total_images / batch_size)
logger.info(
f"Processing {total_images} images in {num_batches} batches of size {batch_size}"
)
for batch_idx in range(num_batches):
start_idx = batch_idx * batch_size
end_idx = min((batch_idx + 1) * batch_size, total_images)
current_batch = image_docs[start_idx:end_idx]
logger.info(
f"Processing batch {batch_idx+1}/{num_batches} with {len(current_batch)} images"
)
# Process each image in the batch (upload to Cloudinary)
batch_image_data = []
for doc in current_batch:
if "image_data" in doc.metadata:
# Create a temporary file for the image
image_id = str(uuid.uuid4())
temp_dir = tempfile.mkdtemp()
# Get the original image
original_img = doc.metadata["image_data"]
# Get original dimensions
width, height = original_img.size
llm_img_size = (128, 128) # Proper size for image processing
if width > llm_img_size[0] or height > llm_img_size[1]:
# Calculate aspect ratio
aspect_ratio = width / height
# Determine new dimensions while preserving aspect ratio
if width > height:
new_width = min(width, llm_img_size[0])
new_height = int(new_width / aspect_ratio)
else:
new_height = min(height, llm_img_size[1])
new_width = int(new_height * aspect_ratio)
# Resize the image
resized_img = original_img.resize(
(new_width, new_height), Image.LANCZOS
)
logger.info(
f"Resized image from {width}x{height} to {new_width}x{new_height}"
)
else:
# Keep original size for smaller images
resized_img = original_img
logger.info(f"Kept original image size: {width}x{height}")
# Save resized image to temporary file
img_path = os.path.join(temp_dir, f"{image_id}.png")
resized_img.save(img_path, format="PNG")
# Convert to base64 for LLM processing
buffered = io.BytesIO()
resized_img.save(buffered, format="PNG")
img_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
base64_url = f"data:image/png;base64,{img_base64}"
# Upload to Cloudinary
upload_result = upload_image(
file_path=img_path,
folder="robokki_images",
public_id=image_id,
)
# Get public URL
public_url = upload_result["secure_url"]
# Store image data for batch processing
batch_image_data.append(
{
"public_url": public_url,
"base64_url": base64_url,
"temp_dir": temp_dir,
"img_path": img_path,
}
)
# Process the batch with LLM (get captions)
batch_inputs = []
for img_data in batch_image_data:
batch_inputs.append(
{
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": "Mô tả hình ảnh này để trích xuất captioning",
},
{
"type": "image_url",
"image_url": {"url": img_data["base64_url"]},
},
],
},
],
"messages_history": [],
}
)
# Get captions for the batch
try:
batch_captions = image_caption_chain.batch(batch_inputs)
# Create document chunks with captions
for i, caption in enumerate(batch_captions):
# Store only the URL in the vector store metadata to avoid size limits
# The base64 data is too large for Pinecone's 40KB metadata limit
processed_image_chunks.append(
Document(
page_content=caption,
metadata={
"type": "image",
"public_url": batch_image_data[i]["public_url"],
},
)
)
# Clean up temporary files
os.remove(batch_image_data[i]["img_path"])
os.rmdir(batch_image_data[i]["temp_dir"])
except Exception as e:
logger.error(f"Error processing batch {batch_idx+1}: {str(e)}")
# Clean up any remaining temporary files
for img_data in batch_image_data:
try:
if os.path.exists(img_data["img_path"]):
os.remove(img_data["img_path"])
if os.path.exists(img_data["temp_dir"]):
os.rmdir(img_data["temp_dir"])
except Exception as cleanup_error:
logger.error(f"Error cleaning up: {str(cleanup_error)}")
raise e
# Process text documents - create a combined text document
combined_text = ""
for doc in text_docs:
if combined_text:
combined_text += "\n\n"
combined_text += doc.page_content
# Chunk the text
text_chunks = []
if combined_text:
# Create a document with the combined text
combined_doc = Document(page_content=combined_text, metadata={"type": "text"})
# Split into chunks
splitter = RecursiveCharacterTextSplitter(
chunk_size=chunk_size, chunk_overlap=chunk_overlap
)
text_chunks = splitter.split_documents([combined_doc])
# Ensure each text chunk has only the 'type' metadata
for chunk in text_chunks:
chunk.metadata = {"type": "text"}
# Combine text chunks and image chunks in the original document order
all_chunks = []
text_idx, image_idx = 0, 0
# Reconstruct the original order based on the input docs
for doc in docs:
if doc.metadata.get("type") == "text":
if text_idx < len(text_chunks):
all_chunks.append(text_chunks[text_idx])
text_idx += 1
elif doc.metadata.get("type") == "image":
if image_idx < len(processed_image_chunks):
all_chunks.append(processed_image_chunks[image_idx])
image_idx += 1
# Add any remaining chunks
all_chunks.extend(text_chunks[text_idx:])
all_chunks.extend(processed_image_chunks[image_idx:])
return all_chunks
def extract_docx_with_images(path: str) -> list[Document]:
"""
Extract text and images from DOCX file.
Args:
path: Path to the DOCX file
Returns:
List of Document objects containing text and images
"""
doc = DocxDoc(path)
docs = []
for para in doc.paragraphs:
text = para.text.strip()
if text:
docs.append(Document(page_content=text, metadata={"type": "text"}))
for rel in doc.part._rels.values():
if "image" in rel.target_ref:
img_data = rel.target_part.blob
image = Image.open(io.BytesIO(img_data))
# Store image data in metadata for later processing
docs.append(
Document(
page_content="", # Will be replaced with caption after processing
metadata={
"type": "image",
"image_data": image,
},
)
)
return docs
def extract_pdf_with_images(pdf_path: str) -> list[Document]:
"""
Extract text and images from PDF.
Args:
pdf_path: Path to the PDF file
Returns:
List of Document objects containing text and images
"""
docs = []
doc = fitz.open(pdf_path)
# Extract text from PDF
for page in doc:
text = page.get_text("text")
if text:
docs.append(Document(page_content=text, metadata={"type": "text"}))
# Extract images from PDF
for img in page.get_images(full=True):
xref = img[0]
base_image = doc.extract_image(xref)
img_bytes = base_image["image"]
# Convert image bytes to PIL Image
image = Image.open(io.BytesIO(img_bytes))
# Store image data in metadata for later processing
docs.append(
Document(
page_content="", # Will be replaced with caption after processing
metadata={
"type": "image",
"image_data": image,
},
)
)
return docs
def process_and_index_file(
file_path: str,
chunk_size: int = 1000,
chunk_overlap: int = 200,
batch_size: int = 30,
bot_id: str = None,
) -> list[Document]:
"""
Process a file and index it in the vector store.
Args:
file_path: Path to the file to process
chunk_size: Size of text chunks
chunk_overlap: Overlap between chunks
batch_size: Number of images to process in a single batch
Returns:
List of processed Document objects
"""
# Process the file
documents = extract_and_chunk_documents(
file_path=file_path,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
upload_images=True,
batch_size=batch_size,
)
# Add bot_id to document metadata if provided
if bot_id:
for doc in documents:
doc.metadata["bot_id"] = bot_id
# Index in vector store
# vector_store_lesson_content.add_documents(documents)
return documents
def process_and_index_directory(
directory_path: str,
file_extensions: list[str] = None,
chunk_size: int = 1000,
chunk_overlap: int = 200,
) -> list[Document]:
"""
Process all files in a directory and index them in the vector store.
Args:
directory_path: Path to the directory
file_extensions: List of file extensions to process (e.g., [".txt", ".md", ".pdf", ".docx"])
chunk_size: Size of text chunks
chunk_overlap: Overlap between chunks
Returns:
List of processed Document objects
"""
all_docs = []
for root, _, files in os.walk(directory_path):
for file in files:
file_path = os.path.join(root, file)
# Skip files with unwanted extensions
if file_extensions and not any(
file.endswith(ext) for ext in file_extensions
):
continue
try:
docs = process_and_index_file(
file_path=file_path,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
)
all_docs.extend(docs)
except Exception as e:
print(f"Error processing {file_path}: {e}")
return all_docs
if __name__ == "__main__":
# Example usage
docs = process_and_index_file("./")
print(f"Processed {len(docs)} chunks")
# Or process a directory
# docs = process_and_index_directory(
# "path/to/your/directory",
# file_extensions=[".txt", ".md", ".pdf", ".docx"]
# )
print(f"Processed {len(docs)} chunks from directory")
pass
|