Daryl Lim
Update app.py
80a5f54
import os
import tempfile
import shutil
import torch
import gradio as gr
from pathlib import Path
from typing import Optional, List, Union
import gc
import time
# Docling imports
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.document_converter import DocumentConverter, PdfFormatOption, WordFormatOption, SimplePipeline
# LangChain imports
from langchain_community.document_loaders import UnstructuredMarkdownLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.schema import Document
# Transformers imports for IBM Granite model
import spaces
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
# Initialize IBM Granite model and tokenizer
print("Loading Granite model and tokenizer...")
model_name = "ibm-granite/granite-3.3-8b-instruct"
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name)
# Create quantization config
quantization_config = BitsAndBytesConfig(
load_in_4bit=True, # Use 4-bit quantization for better memory efficiency
bnb_4bit_compute_dtype=torch.bfloat16 # Use bfloat16 for computation with 4-bit quantization
)
# Load model with optimization for GPU
model = AutoModelForCausalLM.from_pretrained(
model_name,
device_map="auto",
quantization_config=quantization_config
)
print("Model loaded successfully!")
# Helper function to detect document format
def get_document_format(file_path) -> Optional[InputFormat]:
"""Determine the document format based on file extension"""
try:
file_path = str(file_path)
extension = os.path.splitext(file_path)[1].lower()
format_map = {
'.pdf': InputFormat.PDF,
'.docx': InputFormat.DOCX,
'.doc': InputFormat.DOCX,
'.pptx': InputFormat.PPTX,
'.html': InputFormat.HTML,
'.htm': InputFormat.HTML
}
return format_map.get(extension)
except Exception as e:
print(f"Error in get_document_format: {str(e)}")
return None
# Function to convert documents to markdown
def convert_document_to_markdown(doc_path) -> str:
"""Convert document to markdown using simplified pipeline"""
try:
# Convert to absolute path string
input_path = os.path.abspath(str(doc_path))
print(f"Converting document: {doc_path}")
# Create temporary directory for processing
with tempfile.TemporaryDirectory() as temp_dir:
# Copy input file to temp directory
temp_input = os.path.join(temp_dir, os.path.basename(input_path))
shutil.copy2(input_path, temp_input)
# Configure pipeline options
pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = False # Disable OCR for performance
pipeline_options.do_table_structure = True
# Create converter with optimized options
converter = DocumentConverter(
allowed_formats=[
InputFormat.PDF,
InputFormat.DOCX,
InputFormat.HTML,
InputFormat.PPTX,
],
format_options={
InputFormat.PDF: PdfFormatOption(
pipeline_options=pipeline_options,
),
InputFormat.DOCX: WordFormatOption(
pipeline_cls=SimplePipeline
)
}
)
# Convert document
print("Starting conversion...")
conv_result = converter.convert(temp_input)
if not conv_result or not conv_result.document:
raise ValueError(f"Failed to convert document: {doc_path}")
# Export to markdown
print("Exporting to markdown...")
md = conv_result.document.export_to_markdown()
# Create output path
output_dir = os.path.dirname(input_path)
base_name = os.path.splitext(os.path.basename(input_path))[0]
md_path = os.path.join(output_dir, f"{base_name}_converted.md")
# Write markdown file
with open(md_path, "w", encoding="utf-8") as fp:
fp.write(md)
return md_path
except Exception as e:
return f"Error converting document: {str(e)}"
# Improved text processing function
def clean_and_prepare_text(markdown_path):
"""Load, clean and prepare document text for better processing"""
try:
# Load the document
loader = UnstructuredMarkdownLoader(str(markdown_path))
documents = loader.load()
if not documents:
return None, "No content could be extracted from the document."
# Combine all document content for pre-processing
raw_text = " ".join([doc.page_content for doc in documents])
# Clean up the text
# 1. Normalize whitespace
text = " ".join(raw_text.split())
# 2. Fix common OCR and conversion artifacts
text = text.replace(" .", ".").replace(" ,", ",")
# 3. Ensure proper spacing after punctuation
for punct in ['.', '!', '?']:
text = text.replace(f"{punct}", f"{punct} ")
# Split into improved documents
# Use a sensible paragraph size
paragraphs = [p.strip() for p in text.split("\n\n") if p.strip()]
# Create structured documents for better processing
processed_docs = []
for i, para in enumerate(paragraphs):
if len(para) > 10: # Skip very short paragraphs
processed_docs.append(Document(
page_content=para,
metadata={"source": markdown_path, "paragraph": i}
))
return processed_docs, None
except Exception as e:
return None, f"Error processing document text: {str(e)}"
# Improved text splitting configuration
def create_optimized_text_splitter():
"""Create an optimized text splitter for document processing"""
return RecursiveCharacterTextSplitter(
chunk_size=800, # Slightly smaller for more focused chunks
chunk_overlap=150, # Increased overlap to maintain context
length_function=len,
separators=["\n\n", "\n", ".", "!", "?", ";", ":", " ", ""] # More comprehensive separators
)
# Function to generate a summary using the IBM Granite model
def generate_summary(chunks: List[Document], length_type="sentences", length_count=3):
"""Generate a summary from document chunks using the IBM Granite model
Args:
chunks: List of document chunks to summarize
length_type: Either "sentences" or "paragraphs"
length_count: Number of sentences (1-10) or paragraphs (1-3)
"""
# Print debug information
print(f"Generating summary with length_type={length_type}, length_count={length_count}")
# Ensure length_count is an integer
try:
length_count = int(length_count)
except (ValueError, TypeError):
print(f"Failed to convert length_count to int: {length_count}, using default 3")
length_count = 3
# Apply limits based on type
if length_type == "sentences":
length_count = max(1, min(10, length_count)) # Limit to 1-10 sentences
else: # paragraphs
length_count = max(1, min(3, length_count)) # Limit to 1-3 paragraphs
# Clean and concatenate the text from chunks
# Remove any excessive whitespace and normalize
cleaned_chunks = []
for chunk in chunks:
text = chunk.page_content
# Remove excessive newlines and whitespace
text = ' '.join(text.split())
cleaned_chunks.append(text)
combined_text = " ".join(cleaned_chunks)
# More explicit and forceful prompt structure
if length_type == "sentences":
length_instruction = f"Create a concise summary that is EXACTLY {length_count} complete sentences. Not {length_count-1} sentences. Not {length_count+1} sentences. EXACTLY {length_count} sentences."
else: # paragraphs
length_instruction = f"Create a concise summary that is EXACTLY {length_count} paragraphs. Each paragraph should be 2-4 sentences long. Not {length_count-1} paragraphs. Not {length_count+1} paragraphs. EXACTLY {length_count} paragraphs."
# More detailed prompt with examples of what constitutes a sentence
prompt = f"""<instruction>
You are an expert document summarizer. Your task is to create a high-quality summary of the following text.
{length_instruction}
Remember:
- Your summary must capture the main points of the document
- Your summary must be in your own words (not copied text)
- Your summary must be clearly written and well-structured
- Do not include any explanations, headings, bullet points, or additional formatting
- Respond ONLY with the summary text itself
</instruction>
<text>
{combined_text}
</text>
"""
# Calculate appropriate max_new_tokens but with stricter limits
if length_type == "sentences":
# Approximately 20 tokens per sentence
max_tokens = length_count * 40
else: # paragraphs
# Approximately 100 tokens per paragraph
max_tokens = length_count * 150
# Ensure minimum tokens and add buffer
max_tokens = max(100, min(1500, max_tokens))
print(f"Using max_new_tokens={max_tokens}")
# Generate with lower temperature for more consistent results
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
output = model.generate(
**inputs,
max_new_tokens=max_tokens,
temperature=0.3, # Lower temperature for more deterministic output
top_p=0.9,
do_sample=True,
repetition_penalty=1.2 # Discourage repetition
)
# Decode and return the generated summary
summary = tokenizer.decode(output[0], skip_special_tokens=True)
# Extract just the generated response (after the prompt)
summary = summary[len(tokenizer.decode(inputs["input_ids"][0], skip_special_tokens=True)):]
summary = summary.strip()
# Post-process the summary to ensure it meets the length constraints
if length_type == "sentences":
# Simple sentence counting based on periods
sentences = [s.strip() for s in summary.split('.') if s.strip()]
if len(sentences) > length_count:
# Take only the requested number of sentences
summary = '. '.join(sentences[:length_count]) + '.'
elif len(sentences) < length_count:
# If we have too few sentences, log this issue
print(f"Warning: Generated only {len(sentences)} sentences instead of {length_count}")
return summary.strip()
# Function to process document chunks efficiently
def process_document_chunks(texts, batch_size=8):
"""Process document chunks in efficient batches"""
try:
# Create embeddings with optimized settings
embeddings = HuggingFaceEmbeddings(
model_name="nomic-ai/nomic-embed-text-v1",
model_kwargs={'trust_remote_code': True}
)
# Create vector store more efficiently
vectorstore = FAISS.from_documents(
texts,
embeddings,
# Add distance function for better retrieval
distance_strategy="cosine"
)
return vectorstore
except Exception as e:
print(f"Error in document processing: {str(e)}")
# Fallback to basic processing if optimization fails
embeddings = HuggingFaceEmbeddings(
model_name="nomic-ai/nomic-embed-text-v1",
model_kwargs={'trust_remote_code': True}
)
return FAISS.from_documents(texts, embeddings)
# Main function to process document and generate summary
@spaces.GPU
def process_document(
file_obj: Optional[Union[str, tempfile._TemporaryFileWrapper]] = None,
length_type: str = "sentences",
length_count: int = 3,
progress=gr.Progress()
):
"""Process a document file and generate a summary"""
try:
# Process input file
if not file_obj:
return "Please provide a file to summarize."
document_path = file_obj.name if hasattr(file_obj, 'name') else str(file_obj)
# Validate document format
format_type = get_document_format(document_path)
if not format_type:
return "Unsupported file format. Please upload a PDF, DOCX, PPTX, or HTML file."
# Convert document to markdown
progress(0.3, "Converting document to markdown...")
markdown_path = convert_document_to_markdown(document_path)
if markdown_path.startswith("Error"):
return markdown_path
# Clean and prepare the text
progress(0.4, "Processing document text...")
processed_docs, error = clean_and_prepare_text(markdown_path)
if error:
return error
# Split the documents with optimized splitter
text_splitter = create_optimized_text_splitter()
texts = text_splitter.split_documents(processed_docs)
if not texts:
return "No text could be extracted from the document."
# Create vector store with efficient processing
progress(0.6, "Processing document content...")
vectorstore = process_document_chunks(texts)
# Create retriever with optimized settings
retriever = vectorstore.as_retriever(
search_type="similarity",
search_kwargs={"k": 4} # Number of chunks to retrieve
)
# Process chunks in smaller batches for memory efficiency
progress(0.8, "Generating summary...")
all_chunks = []
batch_size = 4 # Smaller batch size for memory efficiency
# Get all document chunks
doc_ids = list(vectorstore.index_to_docstore_id.values())
# Process in smaller batches
for i in range(0, len(doc_ids), batch_size):
batch_ids = doc_ids[i:i+batch_size]
batch_chunks = [vectorstore.docstore.search(doc_id) for doc_id in batch_ids]
all_chunks.extend(batch_chunks)
# Force garbage collection to free memory
gc.collect()
# Sleep briefly to allow memory cleanup
time.sleep(0.1)
# Case 1: Very small documents - use all chunks directly
if len(all_chunks) <= 8:
return generate_summary(
all_chunks,
length_type=length_type.lower(),
length_count=length_count
)
# Case 2: Medium-sized documents - process in one batch
elif len(all_chunks) <= 16:
return generate_summary(
all_chunks[:8], # Use first 8 chunks (usually contains most important info)
length_type=length_type.lower(),
length_count=length_count
)
# Case 3: Large documents - process in multiple batches
else:
# First pass: Generate summaries for each batch
summaries = []
for i in range(0, len(all_chunks), batch_size):
batch = all_chunks[i:i+batch_size]
summary = generate_summary(
batch,
length_type="paragraphs", # Use paragraphs for intermediate summaries
length_count=1 # One paragraph per batch
)
summaries.append(summary)
# Force garbage collection
gc.collect()
# Second pass: Generate final summary from batch summaries
final_summary = generate_summary(
[Document(page_content=s) for s in summaries],
length_type=length_type.lower(),
length_count=length_count
)
return final_summary
except Exception as e:
return f"Error processing document: {str(e)}"
# Create Gradio interface
def create_gradio_interface():
"""Create and launch the Gradio interface"""
with gr.Blocks(title="Granite Document Summarization") as app:
gr.Markdown("# Granite Document Summarization")
gr.Markdown("Upload a document to generate a summary.")
with gr.Row():
with gr.Column(scale=1):
file_input = gr.File(
label="Upload Document (PDF, DOCX, PPTX, HTML)",
file_types=[".pdf", ".docx", ".doc", ".pptx", ".html", ".htm"]
)
with gr.Row():
length_type = gr.Radio(
choices=["Sentences", "Paragraphs"],
value="Sentences",
label="Summary Length Type"
)
with gr.Row():
# Use slider for sentence count (1-10)
sentence_count = gr.Slider(
minimum=1,
maximum=10,
value=3,
step=1,
label="Number of Sentences",
visible=True
)
# Use radio for paragraph count (1-3)
paragraph_count = gr.Radio(
choices=["1", "2", "3"],
value="1",
label="Number of Paragraphs",
visible=False
)
submit_btn = gr.Button("Summarize", variant="primary")
with gr.Column(scale=2):
output = gr.TextArea(
label="Summary",
lines=15,
max_lines=30
)
# Add interactivity to show/hide appropriate count selector
def update_count_visibility(length_type):
is_sentences = length_type == "Sentences"
return [
gr.update(visible=is_sentences), # For sentence_count
gr.update(visible=not is_sentences) # For paragraph_count
]
length_type.change(
fn=update_count_visibility,
inputs=[length_type],
outputs=[sentence_count, paragraph_count]
)
# Function to handle form submission properly
def process_document_wrapper(file, length_type, sentence_count, paragraph_count):
# Convert capitalized length_type to lowercase for processing
length_type_lower = length_type.lower()
print(f"Processing with length_type={length_type}, sentence_count={sentence_count}, paragraph_count={paragraph_count}")
# Determine count based on the selected length type
if length_type_lower == "sentences":
# For sentences, use the slider value directly
try:
count = int(sentence_count)
count = max(1, min(10, count)) # Ensure within range 1-10
print(f"Using sentence count: {count}")
except (ValueError, TypeError):
print(f"Invalid sentence count: {sentence_count}, using default 3")
count = 3
else:
# For paragraphs, convert from string to int if needed
try:
# Check if paragraph_count is a string (from radio button)
if isinstance(paragraph_count, str):
count = int(paragraph_count)
# Check if it's a boolean (from visibility toggle)
elif isinstance(paragraph_count, bool):
count = 1 # Default if boolean
else:
count = int(paragraph_count)
count = max(1, min(3, count)) # Ensure within range 1-3
print(f"Using paragraph count: {count}")
except (ValueError, TypeError):
print(f"Invalid paragraph count: {paragraph_count}, using default 1")
count = 1
return process_document(file, length_type_lower, count)
submit_btn.click(
fn=process_document_wrapper,
inputs=[file_input, length_type, sentence_count, paragraph_count],
outputs=output
)
gr.Markdown("""
## How to use:
1. Upload a document (PDF, DOCX, PPTX, HTML)
2. Choose your summary length preference:
- Number of Sentences (1-10)
- Number of Paragraphs (1-3)
3. Click "Summarize" to process the document
*This application uses the IBM Granite 3.3-8b model to generate summaries.*
""")
return app
# Launch the application
if __name__ == "__main__":
app = create_gradio_interface()
app.launch()