distilroberta-base-goodreads-genres

Model Overview

This model is a fine-tuned version of distilroberta-base for Goodreads book review / genre classification. It is designed to classify book-related text, such as book reviews, descriptions, or summaries, into genre categories.

The model was developed as part of an MLOps assignment using Hugging Face Transformers, Kaggle Notebook, Hugging Face Hub, and Weights & Biases for experiment tracking.

Model Details

• Model name: distilroberta-base-goodreads-genres
• Base model: distilroberta-base
• Model type: Transformer-based sequence classification model
• Task: Text classification
• Domain: Goodreads book reviews / genre classification
• Language: English
• Library: Hugging Face Transformers
• Training platform: Kaggle Notebook
• Experiment tracking: Weights & Biases
• Model repository: https://huggingface.co/sumitp76/distilroberta-base-goodreads-genres

Important Links

• Kaggle Notebook: https://www.kaggle.com/code/sumitpiitj/gr-book-review/edit
• Hugging Face Model: https://huggingface.co/sumitp76/distilroberta-base-goodreads-genres
• W&B Project Dashboard: https://wandb.ai/sumit-k-pal-76-iitj/mlops-assignment2/table?nw=nwusersumitkpal76

Setup Instructions

1. Clone or open the project

The training was performed in a Kaggle Notebook.

Kaggle Notebook:

https://www.kaggle.com/code/sumitpiitj/gr-book-review/edit

2. Install dependencies

Install the required Python libraries:

pip install transformers
pip install datasets
pip install evaluate
pip install accelerate
pip install huggingface_hub
pip install wandb
pip install scikit-learn
pip install pandas
pip install numpy
pip install torch

In Kaggle, many packages may already be installed. If needed, install missing packages inside a notebook cell:

!pip install transformers datasets evaluate accelerate huggingface_hub wandb scikit-learn

3. Set up Huggging Face token

To push the model to Hugging Face Hub, create a Hugging Face access token with Write permission.

In Kaggle:

Go to Add-ons Open Secrets Add your Hugging Face token Save it using the name:

HF_TOKEN

Then load it in the notebook:

from kaggle_secrets import UserSecretsClient
from huggingface_hub import login

user_secrets = UserSecretsClient()
HF_TOKEN = user_secrets.get_secret("HF_TOKEN")

login(token=HF_TOKEN)

4. Set up W&B tracking

Log in to Weights & Biases:

import wandb

wandb.login()

Initialize a W&B run:

wandb.init(
    project="mlops-assignment2",
    name="distilroberta-base-goodreads-genres"
)

W&B dashboard:

https://wandb.ai/sumit-k-pal-76-iitj/mlops-assignment2/table?nw=nwusersumitkpal76

5. Train the model

The model was trained using Hugging Face Trainer.

General training flow:

from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments

model_name = "distilroberta-base"

tokenizer = AutoTokenizer.from_pretrained(model_name)

model = AutoModelForSequenceClassification.from_pretrained(
    model_name,
    num_labels=num_labels
)

Tokenize the dataset:

def tokenize_function(examples):
    return tokenizer(
        examples["text"],
        padding="max_length",
        truncation=True
    )

tokenized_dataset = dataset.map(tokenize_function, batched=True)

Train using Trainer:

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    tokenizer=tokenizer,
    compute_metrics=compute_metrics
)

trainer.train()

6. Evaluate the model

After training, evaluate the model:

results = trainer.evaluate()
print(results)

7. Push the model to Hugging Face Hub

repo_id = "sumitp76/distilroberta-base-goodreads-genres"

model.push_to_hub(repo_id, token=HF_TOKEN)
tokenizer.push_to_hub(repo_id, token=HF_TOKEN)

Model link:

https://huggingface.co/sumitp76/distilroberta-base-goodreads-genres

How to use the model

You can use the model directly with the Hugging Face pipeline.

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="sumitp76/distilroberta-base-goodreads-genres"
)

text = "A young wizard discovers his magical powers and enters a hidden world of adventure."

result = classifier(text)

print(result)

Training Details

Training Platform

The model was trained on Kaggle Notebook.

Platform: Kaggle Notebook link: https://www.kaggle.com/code/sumitpiitj/gr-book-review/edit Framework: Hugging Face Transformers Experiment tracking: Weights & Biases Model hosting: Hugging Face Hub

Base Model

The base model used was:

distilroberta-base

distilroberta-base is a smaller and faster version of RoBERTa. It is suitable for text classification tasks where a balance between performance and efficiency is required.

Preprocessing Steps

The general preprocessing steps included:

1. Loading the Goodreads book review / genre dataset
2. Checking and cleaning missing values
3. Preparing the input text column
4. Encoding genre labels into numeric IDs
5. Splitting the dataset into training and evaluation sets
6. Tokenizing text using the distilroberta-base tokenizer
7. Applying truncation and padding
8. Training the model using Hugging Face Trainer

Training Configuration

Update the values below according to the final notebook settings:

Results

Result Link

Kaggle Notebook: https://www.kaggle.com/code/sumitpiitj/gr-book-review/edit Hugging Face model: https://huggingface.co/sumitp76/distilroberta-base-goodreads-genres W&B dashboard: https://wandb.ai/sumit-k-pal-76-iitj/mlops-assignment2/table?nw=nwusersumitkpal76