Create a Parakeet Managed Inference Job in the web interface
Serve the Parakeet speech-to-text model behind an OpenAI-compatible transcription endpoint from the CosmicAC web interface.
Create a Parakeet Managed Inference Job to serve the speech-to-text model behind an OpenAI-compatible transcription endpoint. The form has six sections, and you click Continue to move from one section to the next. For a description of each field in the form, see Parakeet Managed Inference Job configuration.
Prerequisites
Before you start, make sure that you have the following.
- A running CosmicAC deployment. See Set up CosmicAC.
- Access to the CosmicAC web interface.
Steps
Open the new job form
In the left navigation, click Jobs, and then click New Job.
Select the job type
In the What kind of job? section, select Managed Inference, and then click Continue.
Enter the basics
In the Basics section, enter a Job name. Use at least three lowercase letters, numbers, or hyphens.
In the Tags field, add at least one tag. To add a tag, type it, and then press Enter.
Click Continue.
Select the model
In the Model to serve section, select nvidia/parakeet-tdt-0.6b-v3 in the Model list. CosmicAC prefills the Parakeet configuration from the model's recommended configuration.
Name the endpoint
Under Endpoint, enter an Endpoint name. Use a name that no other Managed Inference Job in the deployment uses, including a failed job that you haven't deleted.
Below the name, Will be reachable at shows the endpoint URL.
Edit the Parakeet configuration (optional)
Under Parakeet configuration, edit the Chunk duration, Chunk overlap, and Max file size values that you want to change. For the limits of each field, see Parakeet Managed Inference Job configuration.
Set the root disk
Under Instance resources, in the Root disk (GB) field, select a preset or enter a value.
Click Continue.
Select the hardware
- In the Hardware section, select a location in the Location list.
- Select a GPU.
- Set the GPU count. Below the count, Recommended for this model shows the model's GPU count. On one node shows the most free GPUs on any single node, and In this region shows the total free GPUs in the location.
- In the CUDA / driver list, select a driver.
- Set Replicas to 1, 2, or 4. Total shows the number of GPUs that the job claims.
- Click Continue.
A Parakeet replica always runs on one node. For more information about how replicas work, see Replicas.
Select the notification events
In the Notifications section, turn off each job lifecycle event that you don't want this job to report. CosmicAC turns on all four events by default. For more information about the events and their delivery, see Set up webhook notifications.
Click Continue.
Review and create the job
In the Review & launch section, check that the section reports Ready to create, and then click Create job. If the section reports issues, click Edit in the row that names the problem. Fix the problem, and then return to the Review & launch section.
If the form shows a Job creation warning, check the values that the warning names. A job with those values can fail to start. To create the job anyway, select Create anyway with this configuration, and then click Create job.
Open the endpoint
In the Provisioning status dialog, click Open job. When the job's status is Active, click the Endpoint tab to view the endpoint URL. To send audio to the endpoint, follow the steps in Transcribe audio.
Help and troubleshooting
Job stuck in Creating or Starting
If a job stays in Creating or Starting, check the status of its KubeVirt virtual machine instance (VMI).
-
Find the job's container ID. The web interface doesn't show the container ID of a Managed Inference Job. Get it with the CosmicAC CLI.
-
In the left navigation, click Jobs, and then click the job.
-
On the Overview tab, in the Specification section, copy the Job ID.
-
To list the job's containers, run the following command.
cosmicac jobs detail <jobId>Replace
<jobId>with the job ID that you copied.
The output lists one container for each replica. Each container shows a Container ID. If the job has more than one replica, do steps 2 and 3 for each container.
-
-
Find the VMI for the container.
CosmicAC creates one VMI for each container and names it
<container-id>-n0. A multi-node job has one VMI per node.From a machine with
kubectlaccess to your Kubernetes cluster, run the following command.kubectl get vmi -n <namespace>Replace
<namespace>with the namespace configured inK8S_NAMESPACE. -
Check the VMI status.
-
If the VMI is not Running, inspect its events.
kubectl describe vmi <container-id>-n0 -n <namespace> -
If the VMI is Running but the job stays in Creating or Starting, cosmicac-wrk-agent-inference cannot reach cosmicac-wrk-server-k8s-nvidia. These two components connect directly, and some cluster network configurations can block the connection.
To route the connection through a relay, see Set up a relay for CosmicAC.
-