Deploy Banking SMS JSON Parser Chatbot

README.md

@@ -4,7 +4,7 @@ emoji: 🏦                                   # Any valid emoji
 colorFrom: blue                             # Card background gradient (start)
 colorTo: green                              # Card background gradient (end)
 sdk: gradio                                 # You created a Gradio Space
-sdk_version: 5.39.0
+sdk_version: 4.44.0
 app_file: app.py                            # The Python file that launches the app
 pinned: false                               # Show/hide on your profile page
 ---

app.py

@@ -1,13 +1,15 @@
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM
-import torch, json, re, textwrap
+import torch
+import json
+import re
 
 # --------------------------------------------------
 # 1. Model loading
 # --------------------------------------------------
 MODEL_ID = "rawsun00001/banking-sms-json-parser-v6-merged"
 
-print("🔄 Loading banking-SMS JSON parser …")
+print("🔄 Loading banking SMS JSON parser model...")
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
@@ -17,42 +19,37 @@ model = AutoModelForCausalLM.from_pretrained(
 
 if tokenizer.pad_token is None:
     tokenizer.pad_token = tokenizer.eos_token
-print("✅ Model ready!")
+print("✅ Model loaded successfully!")
 
 # --------------------------------------------------
-# 2. Core parsing routine
+# 2. Core parsing function
 # --------------------------------------------------
 def parse_banking_sms(raw_text: str) -> dict:
-    """
-    Accept ANY raw SMS/email snippet and return a dict with the 6 fields:
-    date, type, amount, category, last4, is_transaction
-    """
-    # Collapse internal whitespace so odd line-breaks don’t confuse the model
-    sms = " ".join(raw_text.strip().split())
-    prompt = f"{sms}|"
+    """Parse any raw SMS/email text and return structured JSON"""
+    sms_text = " ".join(raw_text.strip().split())
+    prompt = f"{sms_text}|"
 
     inputs = tokenizer(prompt, return_tensors="pt")
     if torch.cuda.is_available():
-        inputs = {k: v.cuda() for k, v in inputs.items()}
+        inputs = {k: v.to("cuda") for k, v in inputs.items()}
 
     with torch.no_grad():
         outputs = model.generate(
             **inputs,
             max_new_tokens=120,
-            do_sample=False,           # deterministic
+            do_sample=False,
             repetition_penalty=1.05,
             pad_token_id=tokenizer.eos_token_id,
             eos_token_id=tokenizer.eos_token_id,
         )
 
     decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    json_chunk = decoded[len(prompt) :].strip()
+    json_part = decoded[len(prompt):].strip()
 
-    # Robust JSON extraction
-    match = re.search(r"\{[^{}]+\}", json_chunk)
-    if match:
+    json_match = re.search(r'\{[^{}]+\}', json_part)
+    if json_match:
         try:
-            parsed = json.loads(match.group())
+            parsed = json.loads(json_match.group())
             return {
                 "date": parsed.get("date"),
                 "type": parsed.get("type"),
@@ -64,86 +61,89 @@ def parse_banking_sms(raw_text: str) -> dict:
         except json.JSONDecodeError:
             pass
 
-    # Default non-transaction template
-    return dict.fromkeys(
-        ["date", "type", "amount", "category", "last4"], None
-    ) | {"is_transaction": False}
+    return {
+        "date": None,
+        "type": None,
+        "amount": None,
+        "category": None,
+        "last4": None,
+        "is_transaction": False
+    }
 
 # --------------------------------------------------
-# 3. Chatbot wrapper
+# 3. Chatbot response handler
 # --------------------------------------------------
 def chatbot_response(message, history):
-    result = parse_banking_sms(message)
-
-    if result["is_transaction"]:
-        md = f"""
-✅ **Transaction detected**
-
-| Field | Value |
-|-------|-------|
-| **Date** | {result['date']} |
-| **Type** | {result['type'] or 'N/A'} |
-| **Amount** | {result['amount']} |
-| **Category** | {result['category']} |
-| **Last-4** | {result['last4']} |
-
-<details>
-<summary>Full JSON</summary>
-
-{textwrap.indent(json.dumps(result, indent=2), '')}
-
-text
-</details>
-"""
-    else:
-        md = f"""
-ℹ️ **Non-transaction message**
-
-The text you provided looks like a promotional / informational SMS.
-
-{textwrap.indent(json.dumps(result, indent=2), '')}
-
-text
-"""
+    """Handle user input and generate chatbot response"""
+    try:
+        result = parse_banking_sms(message)
+
+        if result["is_transaction"]:
+            response = (
+                f"✅ **Transaction Detected!**\n\n"
+                f"📅 **Date:** {result['date']}\n"
+                f"💳 **Type:** {result['type'].title() if result['type'] else 'N/A'}\n"
+                f"💰 **Amount:** {result['amount']}\n"
+                f"🏪 **Category:** {result['category']}\n"
+                f"🔢 **Last 4 Digits:** {result['last4']}\n\n"
+                f"**Full JSON:**\n```json\n{json.dumps(result, indent=2)}\n```"
+            )
+        else:
+            response = (
+                "ℹ️ **Non-Transaction Message**\n\n"
+                "This appears to be a promotional or informational message, not a banking transaction.\n\n"
+                f"**Classification:**\n```json\n{json.dumps(result, indent=2)}\n```"
+            )
+    except Exception as e:
+        response = f"❌ **Error:** Sorry, I couldn't parse that message.\n\nError: {str(e)}"
 
     history = history or []
-    history.append((message, md))
+    history.append((message, response))
     return history, history
 
 # --------------------------------------------------
 # 4. Gradio interface
 # --------------------------------------------------
-with gr.Blocks(
-    theme=gr.themes.Soft(),
-    title="🏦 Banking SMS JSON Parser",
-    css=".gradio-container {max-width: 800px !important; margin: auto !important;}",
-) as demo:
-
-    gr.Markdown(
-        """
-# 🏦 Banking SMS JSON Parser Chatbot
-
-Paste **any** banking SMS or email snippet below — no special prompt
-formatting required. The model returns structured JSON and a friendly
-summary.
-"""
-    )
-
-    chatbot = gr.Chatbot(label="Parser Chat", height=450)
+with gr.Blocks(theme=gr.themes.Soft(), title="🏦 Banking SMS JSON Parser") as demo:
+    gr.Markdown("""
+    # 🏦 Banking SMS JSON Parser Chatbot
+    
+    Paste any banking SMS/email below – no special formatting needed!
+    
+    **Features:**
+    - ✅ Detects real transactions vs promotional messages
+    - ✅ Extracts date, amount, merchant, category, account info
+    - ✅ Works with all Indian & global banking formats
+    """)
+
+    chatbot = gr.Chatbot(label="Banking SMS Parser", height=400)
     msg = gr.Textbox(
-        label="Paste SMS / Email text here",
-        placeholder="e.g. Your A/c XX1234 debited for 5,000 at AMAZON …",
-        lines=3,
-        max_lines=10,
+        label="Paste your banking SMS/email here",
+        placeholder="Example: Your A/c XX1234 debited for 5000 at AMAZON...",
+        lines=3
     )
-    state = gr.State([])
-
-    msg.submit(chatbot_response, [msg, state], [chatbot, state])
-    msg.submit(lambda: "", None, msg)  # clear box after Enter
-    gr.Markdown(
-        "---\n**Model:** `rawsun00001/banking-sms-json-parser-v6-merged` — 100 % accuracy on test set;"
-        " 169 MB FP16"
+    chat_history = gr.State([])
+
+    gr.Examples(
+        examples=[
+            ["Your A/c XX1234 debited for 5000 on 15-Jan-2024 at AMAZON"],
+            ["2500 credited to A/c **9876 on 20-Dec-2023 from PAYROLL"],
+            ["Card **4321 used for 120 at STARBUCKS on 10-Nov-2023"],
+            ["Transaction Alert: 45.99 debited from **2468 at NETFLIX"],
+            ["Your account balance is 5000. Thank you for banking with us."],
+            ["Congratulations! You are eligible for a personal loan up to 50000."]
+        ],
+        inputs=msg,
+        label="Try these example messages:"
     )
 
+    msg.submit(chatbot_response, [msg, chat_history], [chatbot, chat_history])
+    msg.submit(lambda: "", None, msg)  # Clear input
+
+    gr.Markdown("---\n**Model:** `rawsun00001/banking-sms-json-parser-v6-merged`")
+
+# --------------------------------------------------
+# 5. App launcher
+# --------------------------------------------------
 if __name__ == "__main__":
     demo.launch()

requirements.txt

@@ -1,5 +1,5 @@
-huggingface_hub==0.25.2
-transformers==4.36.0
+transformers==4.45.0
 torch==2.1.0
-gradio==5.39.0
+gradio==4.44.0
 accelerate==0.24.0
+huggingface_hub==0.25.2