Local Model Upload
Seamlessly upload locally built models into our platform
You can upload custom-built machine learning models from your local development environment directly into the Clarifai platform. To ensure smooth integration with the Clarifai platform, it's essential that your model adheres to the Triton format, which is our preferred data format for model deployment.
We've designed our platform to provide robust support for widely used tools, such as TensorFlow, PyTorch, and ONNX. This ensures that your preferred development environments and frameworks remain fully compatible with our platform, giving you flexibility and convenience in deploying your models.
The local model upload functionality is exclusively available to Enterprise users.
The local model upload feature provides several benefits, including:
- Create highly specialized and tailored AI solutions that cater to your unique business needs.
- Leverage Clarifai’s robust infrastructure to expand the capabilities of your locally built models. For example, you can link an uploaded model with other models in a workflow and unlock new possibilities.
- Share locally built models, enabling you to showcase your innovations and collaborate with others in the AI community.
The uploaded model will be automatically deployed and ready to be evaluated, fine-tuned and customized, combined with other models and agent operators in a workflow, or used to serve inference requests as it is.
Build the Model
Begin by creating the model you intend to deploy on the Clarifai platform, making sure to follow the guidelines of the Triton format.
Then, compress the model into a ZIP file and host it on a publicly accessible URL.
Click here to access a step-by-step guide for creating your own Triton inference model.
Input and Output Fields
Each model type requires different input and output fields. The table below illustrates the relationship between supported models and their corresponding input and output types.
Click here to learn more about the different input and output fields.
| Type | Input | Output |
|---|---|---|
| multimodal-embedder | image,text | embeddings |
| text-classifier | text | softmax_predictions |
| text-embedder | text | embeddings |
| text-to-image | text | image |
| text-to-text | text | text |
| visual-classifier | image | softmax_predictions |
| visual-detector | image | predicted_bboxes, predicted_labels, predicted_scores |
| visual-embedder | image | embeddings |
| visual-segmenter | image | predicted_mask |
How to Upload a Model
After building the model, you can now upload it to the Clarifai platform.
Step 1: Add the Model
Go to the individual page of an application you own, and select the Models option on the collapsible left sidebar.
On the models listing page, click the Add Model button at the upper-right corner of the page.
On the Add a model window that pops up, select the Upload a Model option and click the Continue button.
You'll be redirected to the Triton Model Upload page, where you can upload your custom-built model.
Step 2: Fill out the Model Upload Form
You can now start filling out the model upload form.
This form is where you'll provide essential information about your model and its specifications, ensuring that it is properly configured and ready for integration into the Clarifai platform.
Let’s talk about the fields to fill in the form.
Model ID
Carefully pick a memorable, human-readable ID for the model you want to upload.
For this example, we'll specify the ID as
dummy-classification-model.
Model Type
Choose a model type from the available options in the drop-down list.
For this example, we'll specify the model type as
visual-classifier.
Clarifai Model Notes
Provide a short description of what the model is about. Later, you can go to the model's individual page, under the Notes section, and edit its description using Markdown.
Model Version Description
Optionally, you can provide a brief description of the model version, summarizing its characteristics or updates.
Triton Model Zip URL
Enter a publicly accessible and downloadable URL for the zipped Triton model.
For this example, we'll enter the URL as
s3://clarifai-testing-persistent-data/test_triton_upload/dummy-visual-classifier.zip. It's a simple, dummy visual-classifier model that returns five concepts, each with an identical confidence score.
Input Fields
Provide the required input parameters for the model.
For this example, we'll specify the input parameter as
image.
Output Fields
Provide the required output parameters for the model.
For this example, we'll specify the output parameter as
softmax_predictions.
Select the Checkbox
Select the checkbox to acknowledge uploading your custom-built model to the Clarifai platform.
Upload Model
Finally, click the Upload Model button to initiate uploading your model.
Step 3: Use the Model
You’ll be redirected to the individual page of the model.
If you check the model's versions table, you’ll notice that the model is still being uploaded. It takes a few minutes for the model to be uploaded — large models could take longer.
After the model has been uploaded, the status will change to Model Trained.
You can now click the Use Model button at the upper-right corner of the page to start using the model.