​
Supported Frameworks

You can use a transformer or a diffuser the based model from Hugging face.

​
Steps to load your model

To import a model from Hugging Face, below are the requirements.

As a next step, we would need to import the Hugging Face model into GitHub before we push it to Inverness. How Inferless works is:Hugging Face-> Copy and create a repo in GitHub -> Load the model repo into Inferless.

You can use the imported GitHub Repo to change the pre-processing and post-processing code.

​
Pre Requisite: Note the Model Name, Type, and Framework

  • Navigate to the Hugging Face model page of your choice that you want to import into Inferless.

  • Take note of the "Model Name" (you can also use the copy button), Task Type, Model Framework, and Model Type. These will be required for the next steps.

The fields to be copied/noted are mentioned in RED.

​
Step 1: Add Model in your workspace.

  • Navigate to your desired workspace in Inferless and Click on "Add a custom model" button that you see on the top right. An import wizard will open up.

Click on Add Model

​
Step 2: Choose the source of your model.

  • Since we are using a model from Hugging Face in this example, select Hugging Face as the method of uploading from the Provider list.

  • To proceed with the upload, you will need to connect your Hugging Face account and a private GitHub account. This is a mandatory step as the Hugging Face credentials enable us to import your private repositories, while GitHub creates the template of the Hugging face repo where you can make changes later to modify your pre-processing and post-processing function

  • For more information: Inferless works by copying the model from Hugging Face and creating a new repository in GitHub. The model repository is then loaded into Inferless.

Select Hugging Face and also add the neccesary accounts

​
Step 3: Enter the model details

  • Model Details : In this step, Add your model name(The name that you wish to call your model), Choose the model type(Eg: Transformer), Choose the task type (Eg: Text generation) and Huggingface model name.

Enter the details as noted

  • In case you would like to set up Automatic rebuild for your model, enable it
    • You would need to set up a webhook for this method. Click here for more details.

​
Step 4: Edit the Inference Code and Input/Output Schema

  • Model Codec: In this step, you can modify the input params ( by adding to input_schema.py ) and output params, you can also modify the model load and inference code in app.py

Enter the details as noted

​
Step 5: Configure Machine and Environment.

  • Choose the type of machine, and specify the minimum and maximum number of replicas for deploying your model.
    • Min scale -
    The number of inference workers to keep on at all times.
    • Max scale -
    The maximum number of inference workers to allow at any point of time
    • Configure Custom Runtime ( If you have pip or apt packages), choose Volume, Secrets and set Environment variables like Inference Timeout / Container Concurrency / Scale Down Timeout

Set runtime and configuration

​
Step 6: Review your model details

  • Once you click “Continue,” you will be able to review the details added for the model.

  • If you would like to make any changes, you can go back and make the changes.

  • Once you have reviewed everything, click Deploy to start the model import process.

Review all the details carefully before proceeding

​
Step 7 : Run your model

  • Once you click submit, the model import process would start.

  • It may take some time to complete the import process, and during this time, you will be redirected to your workspace and can see the status of the import under "In Progress/Failed" tab.

    View the model under `In-Progress/ Failed`

  • If you encounter any errors during the model import process or if you want to view the build logs for any reason, you can click on the three dots menu and select “View build logs”. This will show you a detailed log of the import process, which can help you troubleshoot any issues you may encounter.

  • Post-upload, the model will be available under “My Models”

  • You can then select the model and go to "My Model" -> API -> Inference Endpoint details. Here you would find the API endpoints that can be called. You can click on the copy button on the right and can call your model.

Under the API Tab, you can view the API endpoint details.

​
Extra Step: Getting API key details

  • You can now call using this from your end. The inference result would be the output for these calls.
  • In case you need help with API Keys:
    • Click on settings, available on the top, next to your Workspace Name
    • Click on “Workspace API keys”
    • You can view the details of your key or generate a new one

      Sample for now

Here is a sample video of the whole process for a 7GB Stable diffusion Model : Click to view

​
Using CLI

You can use any model you want from Huggingface and deploy with us

  1. Connect your Huggingface account using this command(below) and enter the name and apikey. Make sure your github account is integrated with Inferless

    inferless integration add HF --name <name> --api-key <apikey>

  2. Once you’re done with integration. Run this command to initialise the model

    inferless init hf --name <modelname> --hfmodelname <hfreponame> --modeltype <transfomer/diffuser> --tasktype <type>

    Transformer options:

    • audio-classification
    • automatic-speech-recognition
    • conversational
    • depth-estimation
    • document-question-answering
    • feature-extraction
    • fill-mask
    • image-classification
    • image-segmentation
    • image-to-text
    • object-detection
    • question-answering
    • summarization
    • table-question-answering
    • text-classification
    • text-generation
    • text2text-generation
    • token-classification
    • translation
    • video-classification
    • visual-question-answering
    • zero-shot-classification
    • zero-shot-image-classification
    • zero-shot-object-detection

    Diffuser options:

    • Depth-to-Image
    • Image-Variation
    • Image-to-Image
    • Inpaint
    • InstructPix2Pix
    • Stable-Diffusion-Latent-Upscaler

  3. Now that your model is initialised.

    • To do default deployment use this command
    inferless deploy --gpu t4
    • To do customised deployment use this command
    inferless deploy --gpu t4 --region <regionname> --runtime <runtimename> --volume <volumename> --fractional
Managing Secrets on InferlessGit (Custom Code)