Create app.py

app.py

@@ -0,0 +1,97 @@
+import fitz  # PyMuPDF
+import gradio as gr
+import torch
+from transformers import AutoConfig, AutoTokenizer, T5ForConditionalGeneration
+
+MODEL_REPO = "CaseLoop/pdf"
+FALLBACK_BASE = "google/flan-t5-base"
+
+# Load Tokenizer
+try:
+  tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, legacy=False)
+except Exception:
+  tokenizer = AutoTokenizer.from_pretrained(FALLBACK_BASE, legacy=False)
+
+# Load Model
+try:
+  model = T5ForConditionalGeneration.from_pretrained(MODEL_REPO)
+except Exception:
+  try:
+    config = AutoConfig.from_pretrained(FALLBACK_BASE)
+    model = T5ForConditionalGeneration.from_pretrained(
+        MODEL_REPO, config=config
+    )
+  except Exception:
+    model = T5ForConditionalGeneration.from_pretrained(FALLBACK_BASE)
+
+model.eval()
+
+
+def extract_chunks(pdf_path: str, chunk_size: int = 180, max_chunks: int = 6):
+  doc = fitz.open(pdf_path)
+  lines = []
+  for page in doc:
+    text = page.get_text("text")
+    lines.extend([
+        line.strip()
+        for line in text.split("\n")
+        if len(line.strip().split()) >= 5
+    ])
+
+  words = " ".join(lines).split()
+  if not words:
+    return []
+  return [
+      " ".join(words[i : i + chunk_size])
+      for i in range(0, len(words), chunk_size)
+  ][:max_chunks]
+
+
+def simplify_document(pdf_file):
+  if pdf_file is None:
+    return "Please upload a valid PDF file."
+
+  chunks = extract_chunks(pdf_file.name)
+  if not chunks:
+    return (
+        "No readable text found in document. It may contain scanned image pages."
+    )
+
+  takeaways = []
+  for idx, chunk in enumerate(chunks, 1):
+    prompt = f"simplify bureaucratic text into simple language: {chunk}"
+    inputs = tokenizer(
+        prompt, return_tensors="pt", max_length=512, truncation=True
+    )
+    with torch.no_grad():
+      output_tokens = model.generate(
+          **inputs,
+          max_length=120,
+          min_length=20,
+          num_beams=4,
+          no_repeat_ngram_size=3,
+          repetition_penalty=2.5,
+          early_stopping=True,
+      )
+    summary = tokenizer.decode(output_tokens[0], skip_special_tokens=True)
+    takeaways.append(f"{idx}. {summary}")
+
+  return "\n\n".join(takeaways)
+
+
+demo = gr.Interface(
+    fn=simplify_document,
+    inputs=gr.File(label="Upload Civic / RTI PDF", file_types=[".pdf"]),
+    outputs=gr.Textbox(
+        label="Plain-Language Takeaways", lines=10, interactive=False
+    ),
+    title="📄 Civic & RTI PDF Simplifier",
+    description=(
+        "Upload complex bureaucratic documents to extract simplified summaries"
+        " using fine-tuned FLAN-T5."
+    ),
+    flagging_mode="never",
+)
+
+if __name__ == "__main__":
+  demo.launch()