was getting negative irrespective of input

app.py

@@ -1,37 +1,35 @@
-from tensorflow import keras
-import tensorflow as tf
-from tensorflow.keras.datasets import imdb
-import numpy as np
 import gradio as gr
+import torch
+import tensorflow as tf
+from transformers import AutoTokenizer
+from model import SentimentClassifier
+
+model_state_dict = tf.keras.load_model('sentimentality.h5')
+model = SentimentClassifier(2)
+model.load_state_dict(model_state_dict)
+model.eval()
+
+tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
 
-number_of_words = 3000
-words_per_view = 30
 
-loaded_model = tf.keras.models.load_model('sentimentality.h5')
-word_to_index = imdb.get_word_index()
+def preprocess(text):
+    inputs = tokenizer(text, padding='max_length',
+                       truncation=True, max_length=512, return_tensors='pt')
+    return inputs
+# Define a function to use the model to make predictions
+def predict(review):
+    inputs = preprocess(review)
+    with torch.no_grad():
+        outputs = model(inputs['input_ids'], inputs['attention_mask'])
+        predicted_class = torch.argmax(outputs[0]).item()
+    if(predicted_class==0):
+        return "It was a negative review"
+    return "It was a positive review"
 
-def get_predict(userInputString, model):
-  words = userInputString.split()
-  #print(len(words))
-  encoded_word = np.zeros(words_per_view).astype(int)
-  encoded_word[words_per_view -len(words) - 1] = 1
-  for i, word in enumerate(words):
-    index = words_per_view - len(words) + i
-    encoded_word[index] = word_to_index.get(word, 0) + 3
-  encoded_word = np.expand_dims(encoded_word, axis=0)
-  prediction = model.predict(encoded_word)
-  return prediction
+# Create a Gradio interface
+input_text = gr.inputs.Textbox(label="Input Text")
+output_text = gr.outputs.Textbox(label="Output Text")
+interface = gr.Interface(fn=predict, inputs=input_text, outputs=output_text)
 
-def analyze_sentiment(userInputString):
-  result = get_predict(userInputString, loaded_model)[0][0]
-  if result > 0.5:
-    answer = 'positive review'
-  else: answer = 'negative review'  
-  return answer
-UserInputPage = gr.Interface(
-    fn=analyze_sentiment,
-    inputs = ["text"],
-    outputs=["text"]
-)
-tabbed_Interface = gr.TabbedInterface([UserInputPage], ["Check user input"])
-tabbed_Interface.launch()
+# Run the interface
+interface.launch()