You can deploy this model and start earning money today!

OmarShamakh/
FoodExtract-gemma-3-270m-fine-tune-v1

Name: FoodExtract-gemma-3-270m-fine-tune-v1
Availability: InStock

generated_from_trainer

sft

trl

conversational

google/gemma-3-270m-it

finetune

gemma

text-generation-inference

endpoints_compatible

Views

No views yet

FoodExtract-gemma-3-270m-fine-tune-v1

This is a food and drink extraction language model fine-tuned from Gemma 3 270M.

Given raw text, it is designed to:

Classify the text into food or drink.
Tag the text with one or more tags (see tags dictionary below).
Extract the edible food-related items as a list.
Extract the edible drink-related items as a list.

Example

Input:

British Breakfast with baked beans, fried eggs, black pudding, sausages, bacon, mushrooms, a cup of tea and toast and fried tomatoes

Output:

food_or_drink: 1
tags: fi, di
foods: British Breakfast, baked beans, fried eggs, black pudding, sausages, bacon, mushrooms, toast, fried tomatoes
drinks: tea

Dataset

Fine-tuned on mrdbourke/FoodExtract-1k, 1400 samples of raw text paired with structured food/drink extraction labels generated by gpt-oss-120b.

Tags dictionary

Tag	Meaning
np	nutrition_panel
il	ingredient_list
me	menu
re	recipe
fi	food_items
di	drink_items
fa	food_advertisement
fp	food_packaging

Training results

Mean token accuracy: ~98.3%
Trained for 3 epochs, batch size 16, learning rate 5e-5
Framework: Hugging Face transformers + trl (SFTTrainer)

Quick start

python

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

MODEL_PATH = "OmarShamakh/FoodExtract-gemma-3-270m-fine-tune-v1"

model = AutoModelForCausalLM.from_pretrained(
    MODEL_PATH, dtype="auto", device_map="auto", attn_implementation="eager"
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

input_text = "A plate with bacon, eggs and toast on it"
input_prompt = pipe.tokenizer.apply_chat_template(
    conversation=[{"role": "user", "content": input_text}],
    tokenize=False,
    add_generation_prompt=True
)
output = pipe(input_prompt, max_new_tokens=256)
print(output[0]["generated_text"][len(input_prompt):])

Training procedure

Fine-tuned using Supervised Fine-Tuning (SFT) with Hugging Face's TRL library.