Uploading FoodExtract demo app.py

app.py

@@ -0,0 +1,67 @@
+
+import time
+import torch
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+
+# Load model
+MODEL_PATH = "Janushi/FoodExtract-gemma-3-270m-fine-tune-v1"
+
+loaded_model = AutoModelForCausalLM.from_pretrained(
+    pretrained_model_name_or_path=MODEL_PATH,
+    dtype="auto",
+    device_map="auto",
+    attn_implementation="eager"
+)
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
+
+loaded_model_pipeline = pipeline(
+    "text-generation",
+    model=loaded_model,
+    tokenizer=tokenizer
+)
+
+def pred_on_text(input_text):
+    start_time = time.time()
+    raw_output = loaded_model_pipeline(
+        text_inputs=[{"role": "user", "content": input_text}],
+        max_new_tokens=256,
+        disable_compile=True
+    )
+    end_time = time.time()
+    total_time = round(end_time - start_time, 4)
+    generated_text = raw_output[0]["generated_text"][1]["content"]
+    return generated_text, raw_output, total_time
+
+description = """Extract food and drink items from text using a fine-tuned Gemma-3-270M.
+Fine-tuned on mrdbourke/FoodExtract-1k dataset.
+
+**Input:** Any text or image caption
+**Output:** Structured food/drink extraction
+
+**Example:**
+- Input: "eggs, bacon and toast with orange juice"
+- Output: food_or_drink: 1, foods: eggs, bacon, toast, drinks: orange juice
+"""
+
+demo = gr.Interface(
+    fn=pred_on_text,
+    inputs=gr.TextArea(lines=4, label="Input Text"),
+    outputs=[
+        gr.TextArea(lines=4, label="Generated Text"),
+        gr.TextArea(lines=7, label="Raw Output"),
+        gr.Number(label="Generation Time (s)")
+    ],
+    title="🍳 BiteSight — Food Extraction with Fine-Tuned Gemma-3-270M",
+    description=description,
+    examples=[
+        ["A plate of grilled tofu, salad with avocado and tomatoes"],
+        ["Indian breakfast with roti, tea and fried potatoes"],
+        ["cheese tacos"],
+        ["A photo of a dog sitting on a beach"]
+    ]
+)
+
+if __name__ == "__main__":
+    demo.launch(share=False)