Tasks and traffic - CoreWeave Docs

Create a task

Select Create task and configure:

Alias: The model string callers send.

W&B project: Optional. It defaults to the alias and stores task traces.

Initial model: Optional routing target for version 1.

Proxy backend: Cloudflare for hosted traffic. Local is available only in environments that explicitly enable it.

If a W&B project with the default alias already exists, choose Use a different W&B project and enter that project explicitly to adopt it.

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

The alias is part of the URL. targets is optional, but setting it here makes version 1 callable as soon as routing is applied.

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 '{
    "project": "ticket-classifier",
    "targets": [
      {"model_ref": "openai/gpt-5.6-sol", "weight": 1}
    ]
  }'

See Create a task for the complete schema.

Connect the application

Keep the normal OpenAI client and change two values:

client = OpenAI(
    base_url="https://proxy.training.wandb.ai/v1",
    api_key=os.environ["WANDB_API_KEY"],
)

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

All normal Chat Completions fields are forwarded. The proxy replaces the task alias with the selected target’s provider model and applies that target’s stored parameter overrides.

Verify useful traces

Before collecting a large dataset, confirm:

successful calls appear in the task’s request chart;

traffic share matches routing weights;

a stable wandb.thread_id groups each conversation;

tool definitions, response formats, and multi-turn messages appear in the recorded request;

failed provider calls are visible for debugging but excluded from training data.

Send the complete message history on every call. Each trace becomes one self-contained training example representing the assistant response produced for that history.

​Create a task

​Connect the application

​Verify useful traces

Create a task

Connect the application

Verify useful traces