diff options
| -rw-r--r-- | src/server/ApiManagers/AssistantManager.ts | 54 | ||||
| -rw-r--r-- | src/server/chunker/pdf_chunker.py | 41 |
2 files changed, 41 insertions, 54 deletions
diff --git a/src/server/ApiManagers/AssistantManager.ts b/src/server/ApiManagers/AssistantManager.ts index cfa95cb4e..4d2068014 100644 --- a/src/server/ApiManagers/AssistantManager.ts +++ b/src/server/ApiManagers/AssistantManager.ts @@ -23,6 +23,7 @@ import { AI_Document } from '../../client/views/nodes/chatbot/types/types'; import { Method } from '../RouteManager'; import { filesDirectory, publicDirectory } from '../SocketData'; import ApiManager, { Registration } from './ApiManager'; +import { getServerPath } from '../../client/util/reportManager/reportManagerUtils'; // Enumeration of directories where different file types are stored export enum Directory { @@ -349,47 +350,16 @@ export default class AssistantManager extends ApiManager { method: Method.GET, subscription: '/getResult/:jobId', secureHandler: async ({ req, res }) => { - const { jobId } = req.params; // Get the job ID from the URL parameters - // Check if the job result is available + const { jobId } = req.params; if (jobResults[jobId]) { const result = jobResults[jobId] as AI_Document & { status: string }; - // If the result contains image or table chunks, save the base64 data as image files if (result.chunks && Array.isArray(result.chunks)) { - await Promise.all( - result.chunks.map(chunk => { - if (chunk.metadata && (chunk.metadata.type === 'image' || chunk.metadata.type === 'table')) { - const files_directory = '/files/chunk_images/'; - const directory = path.join(publicDirectory, files_directory); - - // Ensure the directory exists or create it - if (!fs.existsSync(directory)) { - fs.mkdirSync(directory); - } - - const fileName = path.basename(chunk.metadata.file_path); // Get the file name from the path - const filePath = path.join(directory, fileName); // Create the full file path - - // Check if the chunk contains base64 encoded data - if (chunk.metadata.base64_data) { - // Decode the base64 data and write it to a file - const buffer = Buffer.from(chunk.metadata.base64_data, 'base64'); - fs.promises.writeFile(filePath, buffer).then(() => { - // Update the file path in the chunk's metadata - chunk.metadata.file_path = path.join(files_directory, fileName); - chunk.metadata.base64_data = undefined; // Remove the base64 data from the metadata - }); - } else { - console.warn(`No base64_data found for chunk: ${fileName}`); - } - } - }) - ); result.status = 'completed'; } else { result.status = 'pending'; } - res.json(result); // Send the result back to the client + res.json(result); } else { res.status(202).send({ status: 'pending' }); } @@ -417,7 +387,7 @@ export default class AssistantManager extends ApiManager { // If the chunk is an image or table, read the corresponding file and encode it as base64 if (chunk.metadata.type === 'image' || chunk.metadata.type === 'table') { try { - const filePath = serverPathToFile(Directory.chunk_images, chunk.metadata.file_path); // Get the file path + const filePath = path.join(pathToDirectory(Directory.chunk_images), chunk.metadata.file_path); // Get the file path readFileAsync(filePath).then(imageBuffer => { const base64Image = imageBuffer.toString('base64'); // Convert the image to base64 @@ -549,10 +519,24 @@ function spawnPythonProcess(jobId: string, file_name: string, file_data: string) jobProgress[jobId] = { step: 'Complete', progress: 100 }; } catch (err) { console.error('Error parsing final JSON result:', err); + jobResults[jobId] = { error: 'Failed to parse final result' }; } } else { console.error(`Python process exited with code ${code}`); - jobResults[jobId] = { error: 'Python process failed' }; + // Check if there was an error message in stderr + if (stderrOutput) { + // Try to parse the last line as JSON + const lines = stderrOutput.trim().split('\n'); + const lastLine = lines[lines.length - 1]; + try { + const errorOutput = JSON.parse(lastLine); + jobResults[jobId] = errorOutput; + } catch (err) { + jobResults[jobId] = { error: 'Python process failed' }; + } + } else { + jobResults[jobId] = { error: 'Python process failed' }; + } } }); } diff --git a/src/server/chunker/pdf_chunker.py b/src/server/chunker/pdf_chunker.py index 7a3244fbc..130987343 100644 --- a/src/server/chunker/pdf_chunker.py +++ b/src/server/chunker/pdf_chunker.py @@ -64,13 +64,15 @@ class ElementExtractor: A class that uses a YOLO model to extract tables and images from a PDF page. """ - def __init__(self, output_folder: str): + def __init__(self, output_folder: str, doc_id: str): """ Initializes the ElementExtractor with the output folder for saving images and the YOLO model. :param output_folder: Path to the folder where extracted elements will be saved. """ - self.output_folder = output_folder + self.doc_id = doc_id + self.output_folder = os.path.join(output_folder, doc_id) + os.makedirs(self.output_folder, exist_ok=True) self.model = YOLO('keremberke/yolov8m-table-extraction') # Load YOLO model for table extraction self.model.overrides['conf'] = 0.25 # Set confidence threshold for detection self.model.overrides['iou'] = 0.45 # Set Intersection over Union (IoU) threshold @@ -114,20 +116,18 @@ class ElementExtractor: # Save the full page with the red outline table_filename = f"table_page{page_num + 1}_{idx + 1}.png" + file_path_for_client = f"{self.doc_id}/{table_filename}" table_path = os.path.join(self.output_folder, table_filename) page_with_outline.save(table_path) - # Convert the full-page image with red outline to base64 - base64_data = self.image_to_base64(page_with_outline) - tables.append({ 'metadata': { "type": "table", "location": [x1 / img.width, y1 / img.height, x2 / img.width, y2 / img.height], - "file_path": table_path, + "file_path": file_path_for_client, "start_page": page_num, "end_page": page_num, - "base64_data": base64_data, + "base64_data": self.image_to_base64(page_with_outline) } }) @@ -173,21 +173,19 @@ class ElementExtractor: # Save the full page with the red outline image_filename = f"image_page{page_num + 1}_{img_index + 1}.png" + file_path_for_client = f"{self.doc_id}/{image_filename}" image_path = os.path.join(self.output_folder, image_filename) page_with_outline.save(image_path) - # Convert the full-page image with red outline to base64 - base64_data = self.image_to_base64(page_with_outline) - images.append({ 'metadata': { "type": "image", "location": [x1 / page.rect.width, y1 / page.rect.height, x2 / page.rect.width, y2 / page.rect.height], - "file_path": image_path, + "file_path": file_path_for_client, "start_page": page_num, "end_page": page_num, - "base64_data": base64_data, + "base64_data": self.image_to_base64(image) } }) @@ -269,7 +267,7 @@ class PDFChunker: The main class responsible for chunking PDF files into text and visual elements (tables/images). """ - def __init__(self, output_folder: str = "output", image_batch_size: int = 5) -> None: + def __init__(self, output_folder: str = "output", doc_id: str = '', image_batch_size: int = 5) -> None: """ Initializes the PDFChunker with an output folder and an element extractor for visual elements. @@ -279,7 +277,8 @@ class PDFChunker: self.client = Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY")) # Initialize the Anthropic API client self.output_folder = output_folder self.image_batch_size = image_batch_size # Batch size for image processing - self.element_extractor = ElementExtractor(output_folder) # Initialize the element extractor + self.doc_id = doc_id # Add doc_id + self.element_extractor = ElementExtractor(output_folder, doc_id) async def chunk_pdf(self, file_data: bytes, file_name: str, doc_id: str, job_id: str) -> List[Dict[str, Any]]: """ @@ -364,6 +363,7 @@ class PDFChunker: for j, elem in enumerate(batch, start=1): if j in summaries: elem['metadata']['text'] = re.sub(r'^(Image|Table):\s*', '', summaries[j]) + elem['metadata']['base64_data'] = '' processed_elements.append(elem) progress = ((i // image_batch_size) + 1) / total_batches * 100 # Calculate progress @@ -629,10 +629,11 @@ class PDFChunker: return summaries - except Exception: - #print(f"Error in batch_summarize_images: {str(e)}") - #print("Returning placeholder summaries") - return {number: "Error: No summary available" for number in images} + except Exception as e: + # Print errors to stderr so they don't interfere with JSON output + print(json.dumps({"error": str(e)}), file=sys.stderr) + sys.stderr.flush() + class DocumentType(Enum): """ @@ -688,7 +689,7 @@ class Document: """ Process the document: extract chunks, embed them, and generate a summary. """ - pdf_chunker = PDFChunker(output_folder=self.output_folder) + pdf_chunker = PDFChunker(output_folder=self.output_folder, doc_id=self.doc_id) # Initialize PDFChunker self.chunks = asyncio.run(pdf_chunker.chunk_pdf(self.file_data, self.file_name, self.doc_id, self.job_id)) # Extract chunks self.num_pages = self._get_pdf_pages() # Get the number of pages in the document @@ -823,6 +824,8 @@ def main(): output_folder = sys.argv[4] # Get the output folder from arguments try: + os.makedirs(output_folder, exist_ok=True) + # Decode the base64 file data file_bytes = base64.b64decode(file_data) |