How to publish model on Huggingface ??

If you are new to Huggingface and wants to know how can you publish your model to huggingface , so you are at right place. We will learn today how can you publish your very first fine tuned model to huggingface.

Photo by Fahmi Fakhrudin on Unsplash

Pre-requisite : Make sure you have huggingface account to publish your mode.

I am going to use Google Collab to run all code shown in blog but you can have your own local setup to run.

Lets install required libraries for project.

! pip install huggingface_hub
! pip install transformers
! pip install sentencepiece

Now, in order to publish a model on Hugging Face, I’ll begin by reusing an existing model available on Hugging Face and then proceed with the publishing process.

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
# Download Sample Model from Hugging Face to Publish Again 
tokenizer = AutoTokenizer.from_pretrained("ShashiVish/t5-small-fine-tune-cover-letter")
model = AutoModelForSeq2SeqLM.from_pretrained("ShashiVish/t5-small-fine-tune-cover-letter")

Lets save this model locally and will upload locally save folder directly into huggingface.

# Local Path of Model 
model_path = 't5-fine-tune-save-example'
model.save_pretrained(model_path)

You should see a folder getting created on your local system something like below.

Perfect . Now its time to login into huggingface account using your jupyter notebook and upload your model.

#Logging to HuggingFace
from huggingface_hub import login
login()

It will prompt you for Token to after running above code. You can get token from your huggingface Profile > Setting > Access Token. If you had not created token earlier , create new token with write access.

After successful login , we just need final step where we will specify model repository name to be created on huggingface and publish your very first model.

from huggingface_hub import HfApi

api = HfApi()
model_repo_name = "<Your Profile Name>/t5-fine-tune-save-example"  # Format of Input  <Profile Name > / <Model Repo Name> 

#Create Repo in Hugging Face
api.create_repo(repo_id=model_repo_name)

#Upload Model folder from Local to HuggingFace 
api.upload_folder(
    folder_path=model_path,
    repo_id=model_repo_name
)

# Publish Model Tokenizer on Hugging Face
tokenizer.push_to_hub(model_repo_name)

Congratulations. You have successfully published your very first model on Huggingface.

Keep Learning …!! Keep Sharing ..!!