Instill Cloud enables you to build pipelines specifically for unstructured data.
#Get access to pre-trained ML models
A Model component is an algorithm run on unstructured data to solve a certain AI task.
Deploying and maintaining ML models in a production environment is hard! It often takes more effort and a lot longer to deploy a model to production than to develop it offline.
Therefore, we have deployed a list of pre-trained ML models on Instill Cloud. They are ready for use and are totally free during Open Alpha. On the Instill Cloud console, click Model to access all the pre-deployed ML models for different AI tasks. To get the details of a specific model, click on the model ID and see the Description and Setting sections.
You may find that Import models are disabled in Instill Cloud. It is because we are rapidly refactoring Instill Cloud to offer a more stable user experience.
We will decouple the model deployment feature from VDP soon. One is to make the decoupled model deployment feature into a standalone ModelOps platform Instill Model to provide a more stable user experience. The other is to provide an abstract layer in VDP to access remote models. This will open up a lot of opportunities for VDP to integrate with any model-serving platform.
#Set up your first pipeline
First, follow the Get access to pre-trained ML models to check the details of each model and find the model you want. In case the model is not yet supported, consider reporting to our Discord community.
Click on the Pipeline page on the left sidebar. Since we have not yet set up any pipeline, this page will be empty. To create your first pipeline via VDP Console, click Add new pipeline.
Pipeline is the most important concept in VDP. Check the documentation for more details.
#Step 1: Add an HTTP source
VDP currently supports two sources, HTTP and gRPC for SYNC pipeline. An HTTP source accepts HTTP requests with payloads to be processed by a pipeline.
To set up a Source Connector,
- click the Pipeline mode ▾ drop-down and choose
Sync
, - click the Source type ▾ drop-down and choose HTTP, and
- click Next.
#Step 2: Set up a Model
Click the Online models ▾ drop-down and choose the model you want
#Step 3: Add an HTTP destination
Since we are building a SYNC pipeline, the HTTP
destination is paired automatically with the HTTP
source as we set up for the source. Click Next.
When creating pipelines under SYNC mode, source and destination connectors in VDP must be the same, which means:
- HTTP source → HTTP destination
- gRPC source → gRPC destination
#Step 4: Set up a pipeline
We are almost there! We have created Source
, Model
, and Destination
. The last step is to give this pipeline an ID, and we are ready to go! Just
- give your pipeline a unique ID, e,g., instill-sync,
- add a description (optional), and
- click Set up.
🎉 Ta-da! You have your first VDP pipeline setup. You should see it on the Pipeline page. The green light indicates the pipeline is Active and can be triggered via sending HTTP requests.
Check out the docs to understand all the pipeline states.
#Trigger the pipeline with the API token
VDP automatically generates a dedicated trigger endpoint for each pipeline to process unstructured data.
Click the newly created pipeline on the Pipeline page, you will find further details about this pipeline including the corresponding HTTP request examples to trigger the pipeline at the bottom of the page.
#Create an API token
In order to trigger the pipeline, you will need a valid API token. If you don't yet have a token, Click the Settings on the left sidebar, and navigate to the API tokens page. Just click Create Token and give it an ID.
The tokens are non-expiry. Please keep them in a safe place. If your token is ever compromised, choose that token and click Delete. But be careful when you delete an API token. This operation can not be undone. All your apps using this token will stop working.
#Trigger the pipeline
Simply copy the cURL request in the Trigger section, and pass your API token as a Bearer token in the authorization headers. For example, assuming you have created a pipeline for Image Classification, using cURL, you can trigger the pipeline as follows:
We have standardised the input and output format of each AI task. Check out the supported AI task.