Update app.py

app.py

@@ -1,40 +1,24 @@
+import gradio as gr
+from transformers import pipeline
+import re
+
+# Initialize the pipeline with your custom model
+text_pipe = pipeline("text-classification", model="karalif/myTestModel", return_all_scores=True)
+
 def predict(text):
     greeting_pattern = r"^(Halló|Hæ|Sæl|Góða|Kær|Daginn|Kvöldið|Ágæt|Elsku)"
+    
     greeting_feedback = ""
     
-    # Tokenize input text and prepare for model input
-    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
-    input_ids = inputs['input_ids']
-    attention_mask = inputs['attention_mask']
-
-    # Set model to evaluation mode
-    model.eval()
-
-    # Use the model's forward pass
-    output = model(input_ids, attention_mask=attention_mask)
+    # Adding the input text at the beginning of the response, followed by a new line
+    response = f"Input: {text}\n\n"  # Added this line to include the input in the response
     
-    # Get predictions
-    logits = output.logits
-    predictions = torch.softmax(logits, dim=-1)
-    top_pred_indices = predictions.argmax(dim=-1)
-
-    # Calculate attributions using Integrated Gradients
-    attributions = ig.attribute(inputs=input_ids,
-                                additional_forward_args=(attention_mask,),
-                                target=top_pred_indices)
-
-    # Sum the attributions across the embedding dimension and normalize
-    attributions_sum = attributions.sum(dim=-1).squeeze(0)  # Sum across embedding dim
-    attributions_sum = attributions_sum / torch.norm(attributions_sum)
-
-    # Get the most influential token
-    most_influential_token_index = attributions_sum.argmax().item()
-    most_influential_word = tokenizer.decode([input_ids[0][most_influential_token_index]])
-
-    # Prepare the response
-    response = f"Most influential word: {most_influential_word}\n\n"
-    for i, score in enumerate(predictions[0]):
-        label = model.config.id2label[i]
+    results = text_pipe(text)
+    all_scores = results[0]
+    
+    for result in all_scores:
+        label = result['label']
+        score = result['score']
         response += f"{label}: {score:.3f}\n"
     
     if not re.match(greeting_pattern, text, re.IGNORECASE):
@@ -43,3 +27,23 @@ def predict(text):
     response += greeting_feedback
     
     return response
+
+description_html = """
+    <center>
+        <img src='http://www.ru.is/media/HR_logo_vinstri_transparent.png' width='250' height='auto'>
+    </center>
+    """
+
+gr.Interface(
+    fn=predict,
+    inputs=gr.TextArea(label="Enter text here:"),
+    outputs=gr.TextArea(label="Feedback on text input:"),
+    description=description_html,
+    examples=[
+        ["Það voru vitni að árásinni sem tilkynntu málið til lögreglu sem kom skjótt á vettvang."],
+        ["Ég held þetta sé ekki góður tími fara heimsókn."],
+        ["Sæl og blessuð Kristín, hvað er að frella af þér gamla??"],
+        ["Hver á þenan bússtað? já eða nei."],
+        ["Hafi þau svo látið gólfið þorna vel og síðan flotað það til lagfæringar eftir motturnar."],
+    ],
+).launch()