Uploading FoodExtract demo app.py

README.md

@@ -1,5 +1,5 @@
 ---
-title: FoodExtract Fine-tuned LLM Structued Data Extractor
+title: FoodExtract Fine-tuned LLM Structued Data Extractor v1
 emoji: 📝➡️🍟
 colorFrom: green
 colorTo: blue
@@ -12,13 +12,23 @@ license: apache-2.0
 """
 Fine-tuned Gemma 3 270M to extract food and drink items from raw text.
 
-Input can be any form of real text and output will be a formatted string such as the following:
+Input can be any form of real text (mostly focused on shorter image caption-like texts):
+
+```
+A truly eclectic and mouth-watering feast is laid out on the table, featuring savory favorites like crispy fried chicken, 
+a perfectly seared steak, and loaded tacos, complete with a side of creamy mayonnaise. To balance the heavier mains, 
+a vibrant assortment of fresh fruit sits nearby, including a crisp red apple, a tropical pineapple, and a scattering of 
+sweet cherries. Thirst-quenching options complete this extravagant spread, with a classic iced latte, an earthy matcha latte, 
+and a simple, refreshing glass of milk ready to be enjoyed.
+```
+
+And output will be a formatted string such as the following:
 
 ```
 food_or_drink: 1
 tags: fi, re
-foods: tacos, milk, red apple, pineapple, cherries, fried chicken, steak, mayonnaise
-drinks: iced latte, matcha latte
+foods: tacos,red apple, pineapple, cherries, fried chicken, steak, mayonnaise
+drinks: iced latte, matcha latte, milk
 ```
 
 The tags map to the following items:
@@ -34,5 +44,7 @@ tags_dict = {'np': 'nutrition_panel',
  'fp': 'food_packaging'}
 ```
 
-You can see walkthrough step by step code details at: https://www.learnhuggingface.com/notebooks/hugging_face_llm_full_fine_tune_tutorial 
+* You can see walkthrough step by step code details at: https://www.learnhuggingface.com/notebooks/hugging_face_llm_full_fine_tune_tutorial 
+* See the fine-tuning dataset: https://huggingface.co/datasets/mrdbourke/FoodExtract-1k
+* See the fine-tuned model: https://huggingface.co/mrdbourke/FoodExtract-gemma-3-270m-fine-tune-v1
 """