Create app.py

app.py

@@ -0,0 +1,57 @@
+import gradio as gr
+import re
+from PIL import Image
+from io import BytesIO
+import torch
+
+from transformers import DonutProcessor, VisionEncoderDecoderModel
+
+
+def predict(inp):
+    # Load processor
+    processor = DonutProcessor.from_pretrained("jonathanjordan21/donut_fine_tuning_food_composition_id")
+    
+    # Load model
+    model = VisionEncoderDecoderModel.from_pretrained("jonathanjordan21/donut_fine_tuning_food_composition_id")
+
+    # Define Json Parser
+    def get_komposisi(image_path, image=None):
+        image = Image.open(image_path).convert('RGB') if image== None else image.convert('RGB')
+    
+        task_prompt = "<s_kmpsi>"
+        decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
+    
+        pixel_values = processor(image, return_tensors="pt").pixel_values
+    
+        outputs = model.generate(
+            pixel_values.to(device),
+            decoder_input_ids=decoder_input_ids.to(device),
+            max_length=model.decoder.config.max_position_embeddings,
+            early_stopping=True,
+            pad_token_id=processor.tokenizer.pad_token_id,
+            eos_token_id=processor.tokenizer.eos_token_id,
+            use_cache=True,
+            bad_words_ids=[[processor.tokenizer.unk_token_id]],
+            return_dict_in_generate=True,
+        )
+    
+        sequence1 = processor.batch_decode(outputs.sequences)[0]
+        sequence2 = sequence1.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
+        sequence3 = re.sub(r"<.*?>", "", sequence2, count=1).strip()  # remove first task start token
+    
+        return processor.token2json(sequence3)
+
+    #Generate Output
+    out = get_komposisi("", Image.open(BytesIO(image)))
+    return out
+
+
+def upload_file(files):
+    file_paths = [file.name for file in files]
+    return file_paths
+
+gr.Interface(fn=predict,
+             inputs=gr.Image(type="pil"),
+             outputs="json",
+             examples=["lion.jpg", "cheetah.jpg"]).launch()
+