Error
Input text is too short to summarize. Detected {len(input_text)} characters. Please load text by selecting an example from the dropdown menu or by pasting text into the text box.
Output will appear below:
" AGGREGATE_MODEL = "MBZUAI/LaMini-Flan-T5-783M" # model to use for aggregation # if duplicating space: uncomment this line to adjust the max words os.environ["APP_MAX_WORDS"] = str(8182) # set the max words to 2048 os.environ["APP_OCR_MAX_PAGES"] = str(50) # set the max pages to 40 os.environ["APP_AGG_FORCE_CPU"] = str(1) # force cpu for aggregation aggregator = BatchAggregator( AGGREGATE_MODEL, force_cpu=os.environ.get("APP_AGG_FORCE_CPU", False) ) def aggregate_text( summary_text: str, text_file: gr.inputs.File = None, ) -> str: """ Aggregate the text from the batches. NOTE: you should probably include the BatchAggregator object as a fn arg if using this code :param batches_html: The batches to aggregate, in html format :param text_file: The text file to append the aggregate summary to :return: The aggregate summary in html format """ if summary_text is None or summary_text == SUMMARY_PLACEHOLDER: logging.error("No text provided. Make sure a summary has been generated first.") return "Error: No text provided. Make sure a summary has been generated first." try: extracted_batches = extract_batches(summary_text) except Exception as e: logging.info(summary_text) logging.info(f"the batches html is: {type(summary_text)}") return f"Error: unable to extract batches - check input: {e}" if not extracted_batches: logging.error("unable to extract batches - check input") return "Error: unable to extract batches - check input" out_path = None if text_file is not None: out_path = text_file.name # assuming name attribute stores the file path content_batches = [batch["content"] for batch in extracted_batches] full_summary = aggregator.infer_aggregate(content_batches) # if a path that exists is provided, append the summary with markdown formatting if out_path: out_path = Path(out_path) try: with open(out_path, "a", encoding="utf-8") as f: f.write("\n\n## Aggregate Summary\n\n") f.write( "- This is an instruction-based LLM aggregation of the previous 'summary batches'.\n" ) f.write(f"- Aggregation model: {aggregator.model_name}\n\n") f.write(f"{full_summary}\n\n") logging.info(f"Updated {out_path} with aggregate summary") except Exception as e: logging.error(f"unable to update {out_path} with aggregate summary: {e}") full_summary_html = f"""{full_summary}
Input text was truncated to {max_input_length} words. That's about {100*max_input_length/input_wc:.2f}% of the original text.
Dropping stopwords is set to {predrop_stopwords}. If this is not what you intended, please validate the advanced settings.
Input text is too short to summarize. Detected {len(input_text)} characters. Please load text by selecting an example from the dropdown menu or by pasting text into the text box.
Runtime: {rt} minutes with model: {model_name}
" if msg is not None: html += msg html += "" settings["remove_stopwords"] = predrop_stopwords settings["model_name"] = model_name saved_file = saves_summary(summarize_output=_summaries, outpath=None, **settings) return html, full_summary,scores_out, saved_file def load_single_example_text( example_path: str or Path, max_pages: int = 20, ) -> str: """ load_single_example_text - loads a single example text file :param strorPath example_path: name of the example to load :param int max_pages: the maximum number of pages to load from a PDF :return str: the text of the example """ global name_to_path, ocr_model full_ex_path = name_to_path[example_path] full_ex_path = Path(full_ex_path) if full_ex_path.suffix in [".txt", ".md"]: with open(full_ex_path, "r", encoding="utf-8", errors="ignore") as f: raw_text = f.read() text = clean(raw_text, lower=False) elif full_ex_path.suffix == ".pdf": logging.info(f"Loading PDF file {full_ex_path}") max_pages = int(os.environ.get("APP_OCR_MAX_PAGES", max_pages)) logging.info(f"max_pages set to: {max_pages}") conversion_stats = convert_PDF_to_Text( full_ex_path, ocr_model=ocr_model, max_pages=max_pages, ) text = conversion_stats["converted_text"] else: logging.error(f"Unknown file type {full_ex_path.suffix}") text = "ERROR - check example path" return text # def load_uploaded_file(file_obj, max_pages: int = 20, lower: bool = False) -> str: # """ # load_uploaded_file - loads a file uploaded by the user # :param file_obj (POTENTIALLY list): Gradio file object inside a list # :param int max_pages: the maximum number of pages to load from a PDF # :param bool lower: whether to lowercase the text # :return str: the text of the file # """ # global ocr_model # logger = logging.getLogger(__name__) # # check if mysterious file object is a list # if isinstance(file_obj, list): # file_obj = file_obj[0] # file_path = Path(file_obj.name) # try: # logger.info(f"Loading file:\t{file_path}") # if file_path.suffix in [".txt", ".md"]: # with open(file_path, "r", encoding="utf-8", errors="ignore") as f: # raw_text = f.read() # text = clean(raw_text, lower=lower) # elif file_path.suffix == ".pdf": # logger.info(f"loading a PDF file: {file_path.name}") # max_pages = int(os.environ.get("APP_OCR_MAX_PAGES", max_pages)) # logger.info(f"max_pages is: {max_pages}. Starting conversion...") # conversion_stats = convert_PDF_to_Text( # file_path, # ocr_model=ocr_model, # max_pages=max_pages, # ) # text = conversion_stats["converted_text"] # else: # logger.error(f"Unknown file type:\t{file_path.suffix}") # text = "ERROR - check file - unknown file type. PDF, TXT, and MD are supported." # return text # except Exception as e: # logger.error(f"Trying to load file:\t{file_path},\nerror:\t{e}") # return f"Error: Could not read file {file_path.name}. Make sure it is a PDF, TXT, or MD file." def load_uploaded_file(file_objs, max_pages: int = 20, lower: bool = False) -> str: """ load_uploaded_files - loads multiple files uploaded by the user and concatenates their contents :param file_objs (list): List of Gradio file objects :param int max_pages: the maximum number of pages to load from a PDF :param bool lower: whether to lowercase the text :return str: the concatenated text of all the files """ global ocr_model logger = logging.getLogger(__name__) concatenated_text = "" # Initialize an empty string to concatenate text try: for file_obj in file_objs: file_path = Path(file_obj.name) logger.info(f"Loading file:\t{file_path}") if file_path.suffix in [".txt", ".md"]: with open(file_path, "r", encoding="utf-8", errors="ignore") as f: raw_text = f.read() text = clean(raw_text, lower=lower) elif file_path.suffix == ".pdf": logger.info(f"loading a PDF file: {file_path.name}") max_pages = int(os.environ.get("APP_OCR_MAX_PAGES", max_pages)) logger.info(f"max_pages is: {max_pages}. Starting conversion...") conversion_stats = convert_PDF_to_Text( file_path, ocr_model=ocr_model, max_pages=max_pages, ) text = conversion_stats["converted_text"] else: logger.error(f"Unknown file type:\t{file_path.suffix}") text = f"ERROR - check file - unknown file type. PDF, TXT, and MD are supported." concatenated_text += text # Concatenate text from each file return concatenated_text except Exception as e: logger.error(f"Error: {e}") return f"Error: Could not read one or more files. Make sure they are PDF, TXT, or MD files." def parse_args(): """arguments for the command line interface""" parser = argparse.ArgumentParser( description="Document Summarization - Demo", formatter_class=argparse.ArgumentDefaultsHelpFormatter, epilog="Runs a local-only web UI to summarize documents. pass --share for a public link to share.", ) parser.add_argument( "--share", dest="share", action="store_true", help="Create a public link to share", ) parser.add_argument( "-m", "--model", type=str, default=None, help=f"Add a custom model to the list of models: {pp.pformat(MODEL_OPTIONS, compact=True)}", ) parser.add_argument( "-nb", "--add_beam_option", type=int, default=None, help=f"Add a beam search option to the demo UI options, default: {pp.pformat(BEAM_OPTIONS, compact=True)}", ) parser.add_argument( "-batch", "--token_batch_option", type=int, default=None, help=f"Add a token batch size to the demo UI options, default: {pp.pformat(TOKEN_BATCH_OPTIONS, compact=True)}", ) parser.add_argument( "-max_agg", "-2x", "--aggregator_beam_boost", dest="aggregator_beam_boost", action="store_true", help="Double the number of beams for the aggregator during beam search", ) parser.add_argument( "-level", "--log_level", type=str, default="INFO", choices=["DEBUG", "INFO", "WARNING", "ERROR"], help="Set the logging level", ) return parser.parse_args() if __name__ == "__main__": """main - the main function of the app""" logger = logging.getLogger(__name__) args = parse_args() logger.setLevel(args.log_level) logger.info(f"args: {pp.pformat(args.__dict__, compact=True)}") # add any custom options if args.model is not None: logger.info(f"Adding model {args.model} to the list of models") MODEL_OPTIONS.append(args.model) if args.add_beam_option is not None: logger.info(f"Adding beam search option {args.add_beam_option} to the list") BEAM_OPTIONS.append(args.add_beam_option) if args.token_batch_option is not None: logger.info(f"Adding token batch option {args.token_batch_option} to the list") TOKEN_BATCH_OPTIONS.append(args.token_batch_option) if args.aggregator_beam_boost: logger.info("Doubling aggregator num_beams") _agg_cfg = aggregator.get_generation_config() _agg_cfg["num_beams"] = _agg_cfg["num_beams"] * 2 aggregator.update_generation_config(**_agg_cfg) logger.info("Loading OCR model") with contextlib.redirect_stdout(None): ocr_model = ocr_predictor( "db_resnet50", "crnn_mobilenet_v3_large", pretrained=True, assume_straight_pages=True, ) # load the examples name_to_path = load_example_filenames(_here / "examples") logger.info(f"Loaded {len(name_to_path)} examples") demo = gr.Blocks(title="Document Summarization") _examples = list(name_to_path.keys()) logger.info("Starting app instance") with demo: with gr.Column(): gr.HTML("
#
# Load Inputs & Select Parameters
") gr.Markdown( """Enter/paste text below, or upload a file. Pick a model & adjust params (_optional_), and press **Generate Summary!** """ ) with gr.Row(variant="compact"): # with gr.Column(scale=0.5, variant="compact"): # model_name = gr.Dropdown( # choices=MODEL_OPTIONS, # value=MODEL_OPTIONS[0], # label="Model Name", # ) # num_beams = gr.Radio( # choices=BEAM_OPTIONS, # value=BEAM_OPTIONS[len(BEAM_OPTIONS) // 2], # label="Beam Search: # of Beams", # ) # load_examples_button = gr.Button( # "Load Example in Dropdown", # ) # load_file_button = gr.Button("Upload & Process File") with gr.Column(variant="compact"): # model_name = gr.Dropdown( # choices=MODEL_OPTIONS, # value=MODEL_OPTIONS[0], # label="Model Name", # ) model_name = gr.Dropdown( choices=MODEL_OPTIONS, value=MODAL_VALUES[0], label="Model Name", ) # example_name = gr.Dropdown( # _examples, # label="Examples", # value=random.choice(_examples), # ) uploaded_file = gr.File( label="File Upload", file_count="multiple", file_types=[".txt", ".md", ".pdf"], type="file", ) load_file_button = gr.Button("Upload & Process File",varint="primary") with gr.Row(): input_text = gr.Textbox( lines=4, max_lines=12, label="Text to Summarize", placeholder="Enter text to summarize, the text will be cleaned and truncated on Spaces. Narrative, academic (both papers and lecture transcription), and article text work well. May take a bit to generate depending on the input text :)", ) gr.Markdown("---") with gr.Column(): # gr.Markdown("## Generate Summary") with gr.Row(): summarize_button = gr.Button( "Generate Summary", variant="primary", ) # gr.Markdown( # "_Summarization should take ~1-2 minutes for most settings, but may extend up to 5-10 minutes in some scenarios._" # ) output_text = gr.HTML("Summarization should take ~1-2 minutes for most settings, but may extend up to 5-10 minutes in some scenarios.
") with gr.Column(variant="panel"): gr.Markdown("### Results") with gr.Row(): with gr.Column(variant="compact"): gr.Markdown( "Download the summary as a text file, with parameters and scores." ) text_file = gr.File( label="Download as Text File", file_count="single", type="file", interactive=False, ) with gr.Column(variant="compact"): gr.Markdown( "Scores **roughly** represent the summary quality as a measure of the model's 'confidence'. less-negative numbers (closer to 0) are better." ) summary_scores = gr.Textbox( label="Summary Scores", placeholder="Summary scores will appear here", ) with gr.Column(variant="panel"): gr.Markdown("### **Summary Output**") summary_text = gr.HTML( label="Summary", value="