Update app.py

app.py

@@ -1,24 +1,22 @@
 import gradio as gr
 import numpy as np
 import pandas as pd
-from keras.models import load_model
-from huggingface_hub import hf_hub_download
-from transformers import BertTokenizer
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
 from sklearn.preprocessing import LabelEncoder, StandardScaler
 from nltk.sentiment.vader import SentimentIntensityAnalyzer
 import nltk
-from transformers import TFBertModel
-bert_model = TFBertModel.from_pretrained('bert-base-uncased')
+import torch
 
 # Download VADER lexicon for sentiment analysis
 nltk.download('vader_lexicon')
 
-# Load the model from Hugging Face
-model_path = hf_hub_download(repo_id="xeroISB/ServiceNowMTTR", filename="newIncidentModel.h5")
-model = load_model(model_path)
+# Load model and tokenizer from Hugging Face
+tokenizer = AutoTokenizer.from_pretrained("xeroISB/ServiceNowMTTR")
+model = AutoModelForSequenceClassification.from_pretrained("xeroISB/ServiceNowMTTR")
 
-# Initialize BERT tokenizer
-tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+# Move model to GPU if available
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
 
 # Initialize LabelEncoders
 label_encoders = {
@@ -44,9 +42,9 @@ def preprocess_input(short_description, impact, priority, category, urgency):
     
     short_description = input_data['short_description'].iloc[0].lower()
     
-    # Tokenize text data using BERT tokenizer
-    inputs = tokenizer(short_description, return_tensors='tf', padding='max_length', truncation=True, max_length=50)
-    padded_sequences = np.array(inputs['input_ids'])
+    # Tokenize text data using the new tokenizer
+    inputs = tokenizer(short_description, return_tensors='pt', padding='max_length', truncation=True, max_length=50)
+    inputs = {k: v.to(device) for k, v in inputs.items()}
     
     # Feature engineering: Add sentiment score
     sid = SentimentIntensityAnalyzer()
@@ -57,15 +55,19 @@ def preprocess_input(short_description, impact, priority, category, urgency):
     scaler = StandardScaler()
     scaled_numerical_features = scaler.fit_transform(numerical_features)
 
-    # Prepare the final input features
-    X_input = np.concatenate([padded_sequences, scaled_numerical_features], axis=1)
-    return X_input, input_data['sentiment_score'].iloc[0]
+    return inputs, scaled_numerical_features
 
 # Function to make predictions
 def predict(short_description, impact, priority, category, urgency):
-    X_input, sentiment_score = preprocess_input(short_description, impact, priority, category, urgency)
-    predictions = model.predict(X_input)
-    predicted_label = np.argmax(predictions, axis=1)[0]
+    inputs, scaled_numerical_features = preprocess_input(short_description, impact, priority, category, urgency)
+    
+    # Make prediction
+    with torch.no_grad():
+        outputs = model(**inputs)
+        logits = outputs.logits
+        predicted_label = torch.argmax(logits, axis=1).item()
+    
+    sentiment_score = scaled_numerical_features[0][-1]
     return predicted_label, sentiment_score
 
 # Define Gradio interface