distilbert/distilbert-base-uncased-finetuned-sst-2-english

app.py

@@ -2,7 +2,7 @@ import gradio as gr
 from transformers import pipeline
 import re
 
-def anomalies_detector(logs: str) -> list[tuple[int, str]]:
+def anomalies_detector(logs: str) -> list[str]:
     """
     Detect anomalies in software logs using a Hugging Face transformer model.
     This function uses a specialized model trained to identify unusual patterns
@@ -19,12 +19,10 @@ def anomalies_detector(logs: str) -> list[tuple[int, str]]:
     Returns:
         list[tuple[int, str]]: List of tuples containing (line_number, anomalous_text)
     """
-    # Initialize the text classification pipeline with a smaller, more reliable model
-    classifier = pipeline(
-        "text-classification",
-        model="distilbert-base-uncased",  # Using a smaller, more reliable model
-        top_k=2  # Get both normal and anomalous probabilities
-    )
+    # Initialize the text classification pipeline with a proper classification model
+    classifier = pipeline("text-classification",
+                          model="distilbert/distilbert-base-uncased-finetuned-sst-2-english")
+
     
     # Split logs into lines
     log_lines = logs.split('\n')
@@ -38,13 +36,9 @@ def anomalies_detector(logs: str) -> list[tuple[int, str]]:
         # Get classification result
         results = classifier(line)
         
-        # Check if the line is classified as anomalous
-        # The model returns probabilities for both classes
-        for result in results:
-            if result['label'] == 'LABEL_1' and result['score'] > 0.7:  # LABEL_1 indicates potential anomaly
-                anomalies.append((line_num, line))
-                break
-    
+        
+        for log, res in zip(logs, results):
+            anomalies.append(f"{log} => {res}")
     return anomalies
 
 # Create a standard Gradio interface