What is a deployment shape, and why did my deployment fail to create?

A deployment shape is a pre-validated, pre-configured deployment template. It bundles a known-good combination of GPU type and count, precision, and serving parameters, optimized for speed (fast), cost per token at scale (throughput), or lowest cost (minimal). A deployment created from a shape starts from a configuration that is known to work for that model.

Why deployments created without a shape often fail

If you create a deployment without a shape — that is, without passing --deployment-shape (or deploymentShape in the API) — the configuration is not validated ahead of time. Mistakes only surface at creation, where they cause failures. Common examples:

• A GPU count that cannot fit the model in memory
• An accelerator type the model isn’t validated on (for example, requesting H200 for a model whose shapes are all B200)
• A context length the configuration can’t serve
• A quantization or precision the model doesn’t support on that hardware

Deployments created without a shape fail far more often than deployments created from a shape — they are the most common cause of failed deployment creations on Fireworks. Do not create deployments without a shape; the unshaped path may be deprecated in the future.

How to find and use a shape

The shape list is also the authoritative way to discover which GPU types, GPU counts, and precisions a model supports — a hardware combination with no shape is not a validated configuration.

firectl

Match the shapes available for your model:

firectl deployment-shape-version match --model accounts/fireworks/models/gpt-oss-120b

Then pass the shape’s name to --deployment-shape when creating the deployment:

firectl deployment create accounts/fireworks/models/gpt-oss-120b \
  --deployment-shape accounts/fireworks/deploymentShapes/gpt-oss-120b-fast

REST API

Call Match Deployment Shape Versions with a deployment create request for your model — the same match the firectl deployment-shape-version match command runs. It returns the validated shape versions compatible with that model, applying the server-side compatibility rules for you — for LoRA addons and live-merge models, shapes are matched against the model’s base model. To get shapes filtered for LoRA addon serving (enableAddons), include it in the request. The CLI command does not take addons into account:

# YOUR_ACCOUNT_ID is the account that will own the deployment, not the
# model's publisher. The model in the body can live anywhere you can
# deploy it, e.g. accounts/fireworks.
curl -X POST "https://api.fireworks.ai/v1/accounts/YOUR_ACCOUNT_ID/deploymentShapeVersions:match" \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "createDeploymentRequest": {
      "deployment": {
        "baseModel": "accounts/fireworks/models/gpt-oss-120b"
      }
    }
  }'

Then pass one of the returned shape versions as deploymentShape in the Create Deployment request body:

curl -X POST "https://api.fireworks.ai/v1/accounts/YOUR_ACCOUNT_ID/deployments" \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "baseModel": "accounts/fireworks/models/gpt-oss-120b",
    "deploymentShape": "accounts/fireworks/deploymentShapes/gpt-oss-120b-fast"
  }'

Web UI

On the model page, choose Deploy and pick a shape.

If no shape fits

If you need a configuration that no existing shape covers, contact us — we’ll help you find the right shape or add one for your workload.