This page covers multi-tenant Forge at forge.coreweave.com. For W&B Dedicated Cloud or W&B Self-Managed, see Deployment options.
Before you begin
The walkthrough on this page uses Forge in the browser, so you don’t need to install a local tool. To complete it, you need:- A Forge account. If you already have one, sign in to Forge.
- Access to Serverless Inference through included credits or active pay-as-you-go billing. See Serverless Inference usage and limits.
- A Forge project for attributing usage and storing the Weave trace. If you don’t select a project, Serverless Inference uses your default team and the project name
inference.
Accounts and credentials
Forge and CoreWeave Infrastructure use separate identity systems. Forge accounts govern Forge organizations, teams, projects, and product access. CoreWeave accounts govern Infrastructure resources such as CKS, virtual private clouds (VPCs), and CoreWeave AI Object Storage. Some products support both account types. For example, CoreWeave Sandboxes accepts a Forge API key or a CoreWeave API access token for serverless placement. Sandboxes on CKS uses a CoreWeave API access token. For sign-in and administration details, see Forge and CoreWeave identity. If you use Weights & Biases on Multi-tenant Cloud, sign in with your existing account. Your existing projects, runs, and artifacts are available in Forge. For account, application, documentation, API, and URL changes, see What moved in Forge.Tools for your workflow
Forge products provide product UIs, software development kits (SDKs), APIs, and command-line tools for their workflows. Use the documentation for your chosen product to find its installation and authentication requirements.- To connect coding agents and AI assistants to Weights & Biases workflows and data, see Use Weights & Biases with AI assistants.
- To work with CKS and other CoreWeave Infrastructure resources, see CoreWeave developer tools.
Try Forge: Run and trace a model
Use the Serverless Inference Playground to send a request to a hosted model, then inspect the request’s trace in W&B Weave.- Open the Serverless Inference Playground.
-
In the Playground page title bar:
- Click the Select project () button, then select the project to attribute usage to.
- Click the model selector, then select a hosted model.
- Click Trace calls with Weave.
-
In the message box, enter the following prompt:
- Click the Send message button.
- Under the model’s response, click View call to open its trace in Weave.
Build a connected workflow
These tutorials show how data and workloads move between Forge products.Post-train and serve with managed services
Use public-preview Serverless SFT to fine-tune a model, store its adapter in Weights & Biases, and run each checkpoint with Serverless Inference.
Bring a locally trained LoRA to Forge
Train a LoRA locally, upload it to Weights & Biases, and run it through the Serverless Inference Playground or API.
Run a coding agent with Forge
Run OpenCode in a CoreWeave sandbox and use Serverless Inference as its model provider.
Other ways to use Forge
Explore other common Forge workflows and choose the product and first step that match your goal.Train and improve models
Custom training jobs can run on CoreWeave Kubernetes Service (CKS), SUNK, or another environment, while Weights & Biases tracks and compares their results.
Build and evaluate LLM agents and applications
For a comparison of the agent workflows, see Agent Lens and W&B Weave.
Run models and isolated workloads
For dedicated model deployments, compare CoreWeave Inference deployment options. To compare the complete Forge product catalog, see Forge Home.