Update app.py

app.py

@@ -17,10 +17,18 @@ repo_id = "islasher/mbart-spanishToQuechua"
 
 # Cargar el modelo y el tokenizador
 nombre_modelo = 'islasher/mbart-spanishToQuechua'
-model = AutoModelForSeq2SeqLM.from_pretrained(nombre_modelo)
-tokenizer = AutoTokenizer.from_pretrained(nombre_modelo)
 
+#tokenizer = AutoTokenizer.from_pretrained(nombre_modelo)
 
+model_checkpoint = "facebook/mbart-large-50"
+tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_checkpoint)
+
+
+
+from transformers import DataCollatorForSeq2Seq
+
+data_collator = DataCollatorForSeq2Seq(tokenizer) #para preparar los datos
 
 
 
@@ -57,15 +65,19 @@ def compute_metrics(eval_preds):
 
 
 
+from transformers import pipeline
+neutralizer = pipeline('text2text-generation', model='islasher/mbart-spanishToQuechua')
+
+
 
 #CAMBIAR LO QUE SE RETORNA Y PONER LO DEL DECODER.
 
 
-def predict(frase):
-    #img = PILImage.create(img)
-    inputs = tokenizer(frase, return_tensors="pt")
-    outputs = model(**inputs)
-    trad = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    return trad
+# def predict(frase):
+#     inputs = tokenizer(frase, return_tensors="pt")
+#     outputs = model(**inputs)
+#     trad = tokenizer.decode(outputs[0], skip_special_tokens=True)
+#     return trad
+
 # Creamos la interfaz y la lanzamos. 
-gr.Interface(fn=predict, inputs="text", outputs="text").launch(share=False)
+gr.Interface(fn=neutralizer, inputs="text", outputs="text").launch(share=False)