Quick Start - CoreWeave Docs

This guide shows the complete path from an existing AI feature to a specialized model running in production. Open Model Distillation Studio, sign in with your W&B API key, and choose the team that owns the application.

Use a team service-account key for a production application. It avoids coupling availability to one employee’s account.

Step 2: Register your current model provider

Open Providers and connect the service that runs your current model. This lets Model Distillation keep serving the same model while you collect data and train alternatives. You can skip this step when using the built-in wandb-inference provider. The provider must expose an OpenAI-compatible Chat Completions endpoint for routed application traffic. Enter the exact API base URL, including any required path prefix such as /v1; the proxy appends /chat/completions. For Studio relabeling and evaluations, the provider named openai uses OpenAI’s Responses API instead. After saving the provider, add or enable the model ID if it is not already available. Wait until the provider deployment is Applied at the current revision before continuing.

Step 3: Create a task

From Tasks, select Create task. Give the feature a stable alias, such as ticket-classifier, and select the model that serves it today. The task alias becomes the model value used by your application. It stays the same as you train and deploy new implementations.

API: Create a task (PUT /tasks/{alias})

Set the task alias in the URL and the current model in targets:

curl --request PUT \
  --url "https://distillation.training.wandb.ai/v1/tasks/ticket-classifier" \
  --header "Authorization: Bearer $WANDB_API_KEY" \
  --header "Wandb-Entity: your-team" \
  --header "Content-Type: application/json" \
  --data '{
    "targets": [
      {"model_ref": "openai/gpt-5.6-sol", "weight": 1}
    ]
  }'

See Create a task for the complete request and response.

Task routing is published asynchronously. Open the task’s Routing page and wait until the deployment is Applied at the current routing revision before sending application traffic.

Step 4: Connect your application

Keep your normal OpenAI client, but change its base URL and model:

from openai import OpenAI

client = OpenAI(
    base_url="https://proxy.training.wandb.ai/v1",
    api_key=wandb_api_key,
)

response = client.chat.completions.create(
    model="ticket-classifier",
    messages=messages,
    extra_body={
        "metadata": {
            "wandb.entity": "your-team",
            "wandb.thread_id": conversation_id,
        }
    },
)

Model Distillation now collects useful examples from normal application traffic. See Connect your application for Python, JavaScript, and cURL examples.

Step 5: Create a dataset

After the task has representative traffic, create a dataset. Review the examples, remove poor ones, and improve the answers if you want the new model to learn better behavior than the current model provides. Follow the Datasets Quick Start to create and inspect your first dataset.

Step 6: Fine-tune several models

Train one or more candidate models from the dataset. Trying several candidates gives you a better chance of finding the right balance of quality, speed, and cost. Follow the Fine-tuning Quick Start to start a training job.

Step 7: Evaluate the candidates

Compare every candidate with the answers you want to preserve. Review the overall result and individual examples before choosing a winner. Follow the Evaluations Quick Start to run your first comparison.

Step 8: Deploy the winner

Add the winner to task routing. You can start with a small share of traffic, increase it gradually, and keep the original model ready for rollback. Once you trust the workflow, Automation can run the next improvement cycle and deploy a winner only when it meets your quality goal.

Train and deploy your first model

Follow the complete manual workflow with more detail about each decision.

​Step 1: Sign in with W&B

​Step 2: Register your current model provider

​Step 3: Create a task

​Step 4: Connect your application

​Step 5: Create a dataset

​Step 6: Fine-tune several models

​Step 7: Evaluate the candidates

​Step 8: Deploy the winner

Train and deploy your first model

Step 1: Sign in with W&B

Step 2: Register your current model provider

Step 3: Create a task

Step 4: Connect your application

Step 5: Create a dataset

Step 6: Fine-tune several models

Step 7: Evaluate the candidates

Step 8: Deploy the winner