id-recigen-bart

Indonesia Recipe Ingredients Generator Model

WARNING: inference on Huggingface might not run since the tokenizer used is not transformers's tokenizer.

Feel free to test the model in this space

😎 Have fun on generating ingredients 😎

This is a fine-tuned model to generate the Indonesian food ingredients. One of my personal project that I did in my free time.

Basically, you give the name of the food and it will produce the ingredients of the food.

Model

Data: Indonesian Recipe Data on Kaggle
Pre-trained Model: IndoBART-v2

How to use

We will specify the usage of the tokenizer and the model.

Tokenizer

Since we use indobart-v2, we need to use their tokenizer.

First, install the tokenizer by doing pip install indobenchmark-toolkit.

After that, you can load the tokenizer:

from indobenchmark.tokenization_indonlg import IndoNLGTokenizer

tokenizer = IndoNLGTokenizer.from_pretrained("haryoaw/id-recigen-bart")

EDIT:

Seems like the tokenizer in the package is not the same as the one that I use to finetune the model. There are some noticeable bug such as some subword tokens are not considered as subword. Nevertheless, it stil works!

Model

The model can be loaded by using AutoModel.

from transformers import AutoModelForSeq2SeqLM

model = AutoModelForSeq2SeqLM.from_pretrained("haryoaw/id-recigen-bart")

Input Example

Make sure to input a LOWERCASE food name. The tokenizer is case-sensitive!

sayur asam

nasi goreng ayam

~To be continued..