Update app.py

app.py

@@ -1,22 +1,38 @@
 import gradio as gr
-from transformers import pipeline
+import torch
+from transformers import BertForMaskedLM, BertTokenizer
 
-# Laden des Modells für Masked Language Modeling
-unmasker = pipeline('fill-mask', model='bert-base-uncased')
+# Modell und Tokenizer laden mit force_download=True
+model_name = "bert-base-uncased"
+model = BertForMaskedLM.from_pretrained(model_name, force_download=True)
+tokenizer = BertTokenizer.from_pretrained(model_name, force_download=True)
 
-# Gradio Interface
-def masked_language_modeling(text):
-    results = unmasker(text)
-    return results[0]['sequence']
+# Inferenz-Funktion definieren
+def inference(input_text):
+    if "[MASK]" not in input_text:
+        return "Error: The input text must contain the [MASK] token."
+    
+    # Tokenisierung
+    inputs = tokenizer(input_text, return_tensors="pt")
+    mask_token_index = torch.where(inputs["input_ids"] == tokenizer.mask_token_id)[1]
 
+    # Vorhersage
+    with torch.no_grad():
+        outputs = model(**inputs)
+        logits = outputs.logits
+
+    # Wahrscheinlichsten Token für [MASK] finden
+    mask_token_logits = logits[0, mask_token_index, :]
+    top_token = torch.topk(mask_token_logits, 1, dim=1).indices[0].tolist()
+
+    # Vorhersage in den Text einfügen
+    predicted_token = tokenizer.decode(top_token)
+    result_text = input_text.replace("[MASK]", predicted_token, 1)
+    
+    return result_text
+
+# Gradio Interface definieren
 iface = gr.Interface(
-    fn=masked_language_modeling,
-    inputs=gr.Textbox(),
-    outputs=gr.Textbox(),
-    title='BERT Masked Language Modeling',
-    description='Enter a sentence with a [MASK] and see the predictions.'
-)
+    fn=inference,
 
-# Starte die Gradio Benutzeroberfläche
-if __name__ == '__main__':
-    iface.launch()
+    iface.launch(server_port=7862)