Let input text in pipeline

pipeline.py

@@ -43,12 +43,12 @@ class PreTrainedPipeline():
                 - "feature_vector": A list of floats corresponding to the image embedding.
         """
         inputs = data.pop("inputs", data)
-        parameters = data.pop("parameters", {"mode": "image"})
+        parameters = data.pop("parameters", {"mode": "multimodal"})
         
         # decode base64 image to PIL
         image = Image.open(BytesIO(base64.b64decode(inputs['image'])))
         image = self.transform(image).unsqueeze(0).to(device)   
-        text=""
+        text = inputs['text'] if 'text' in inputs else ''# already gets tokenised in the model
         with torch.no_grad():
             feature_vector = self.model(image, text, mode=parameters["mode"])[0,0].tolist()
         # postprocess the prediction