Deployments Quickstart

Deploy models on dedicated GPUs in minutes

On-demand deployments are dedicated GPUs that give you better performance, no rate limits, fast autoscaling, and a wider selection of models than serverless. This quickstart will help you spin up your first on-demand deployment in minutes.

Step 1: Create and export an API key

Before you begin, create an API key in the Fireworks dashboard. Click Create API key and store it in a safe location.

Once you have your API key, export it as an environment variable in your terminal:

macOS / Linux

export FIREWORKS_API_KEY="your_api_key_here"

Windows

setx FIREWORKS_API_KEY "your_api_key_here"

Step 2: Install the CLI

To create and manage on-demand deployments, you’ll need the firectl CLI tool. Install it using one of the following methods, based on your platform:

homebrew

brew tap fw-ai/firectl
brew install firectl

# If you encounter a failed SHA256 check, try first running
brew update

Then, sign in:

firectl signin

Step 3: Create a deployment

This command will create a deployment of GPT OSS 120B. It will take a few minutes to complete. The resulting deployment will scale up to 1 replica.

Requires firectl 1.8.8 or newer — on older versions, default is treated as a literal shape name and the command fails. Check with firectl version.

firectl deployment create accounts/fireworks/models/gpt-oss-120b \
        --deployment-shape default \
        --scale-down-window 5m \
        --scale-up-window 30s \
        --min-replica-count 0 \
        --max-replica-count 1 \
        --scale-to-zero-window 5m \
        --wait

A deployment shape is a pre-configured deployment template created by the Fireworks team that sets sensible defaults for most deployment options (such as hardware type).

--deployment-shape default tells Fireworks to pick a validated shape for the model. To optimize for lower cost-per-token at scale or the lowest cost at small scale, run firectl deployment-shape-version match --model <model-id> and pass a shape with preset type THROUGHPUT or MINIMAL instead — pass the shape’s full resource name to --deployment-shape; preset types are not shape identifier aliases.

Do not omit --deployment-shape: deployments created without a shape do not start from a validated configuration and are much more likely to fail at creation. If no shape fits your workload, contact us; if you deliberately need a shapeless deployment, pass --accept-shapeless-risk (advanced users only). See Deployment shapes and Explicitly creating a deployment without a shape in the deployment guide.

The response will look like this:

Name: accounts/<YOUR ACCOUNT ID>/deployments/<DEPLOYMENT ID>
Create Time: <CREATION_TIME>
Expire Time: <EXPIRATION_TIME>
Created By: <YOUR EMAIL>
State: CREATING
Status: OK
Min Replica Count: 0
Max Replica Count: 1
Desired Replica Count: 0
Replica Count: 0
Autoscaling Policy:
  Scale Up Window: 30s
  Scale Down Window: 5m0s
  Scale To Zero Window: 5m0s
Base Model: accounts/fireworks/models/gpt-oss-120b
...other fields...

Take note of the Name: field in the response, as it will be used in the next step to query your deployment.

Learn more about deployment options→

Learn more about autoscaling options→

Step 4: Query your deployment

Now you can query your on-demand deployment using the same API as serverless models, but using your dedicated deployment. Replace <DEPLOYMENT_NAME> in the below snippets with the value from the Name: field in the previous step:

Python (Fireworks SDK)

Install the Fireworks Python SDK:

The SDK is currently in alpha. Use the --pre flag when installing to get the latest version.

pip

pip install --pre fireworks-ai

Then make your first on-demand API call:

from fireworks import Fireworks

client = Fireworks()

response = client.chat.completions.create(
    model="accounts/fireworks/models/gpt-oss-120b#<DEPLOYMENT_NAME>",
    messages=[{
        "role": "user",
        "content": "Explain quantum computing in simple terms",
    }],
)

print(response.choices[0].message.content)

Python (OpenAI SDK)

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("FIREWORKS_API_KEY"),
    base_url="https://api.fireworks.ai/inference/v1"
)

response = client.chat.completions.create(
    model="<DEPLOYMENT_NAME>",
    messages=[{
        "role": "user",
        "content": "Explain quantum computing in simple terms",
    }],
)

print(response.choices[0].message.content)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.FIREWORKS_API_KEY,
  baseURL: "https://api.fireworks.ai/inference/v1",
});

const response = await client.chat.completions.create({
  model: "<DEPLOYMENT_NAME>",
  messages: [
    {
      role: "user",
      content: "Explain quantum computing in simple terms",
    },
  ],
});

console.log(response.choices[0].message.content);

curl

curl https://api.fireworks.ai/inference/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -d '{
    "model": "<DEPLOYMENT_NAME>",
    "messages": [
      {
        "role": "user",
        "content": "Explain quantum computing in simple terms"
      }
    ]
  }'

The examples from the Serverless quickstart will work with this deployment as well, just replace the model string with the deployment-specific model string from above.

Serverless quickstart→

Common use cases

Autoscale based on requests per second

firectl deployment create accounts/fireworks/models/gpt-oss-120b \
        --deployment-shape default \
        --scale-down-window 5m \
        --scale-up-window 30s \
        --scale-to-zero-window 5m \
        --min-replica-count 0 \
        --max-replica-count 4 \
        --load-targets requests_per_second=5 \
        --wait

Autoscale based on concurrent requests

firectl deployment create accounts/fireworks/models/gpt-oss-120b \
        --deployment-shape default \
        --scale-down-window 5m \
        --scale-up-window 30s \
        --scale-to-zero-window 5m \
        --min-replica-count 0 \
        --max-replica-count 4 \
        --load-targets concurrent_requests=5 \
        --wait

Next steps

Ready to scale to production, explore other modalities, or customize your models?

Upload a custom model

Bring your own model and deploy it on Fireworks

Train Models

Improve model quality with supervised and reinforcement learning

Embeddings & Reranking

Use embeddings & reranking in search & context retrieval

Batch Inference

Run async inference jobs at scale, faster and cheaper

Browse 100+ Models

Explore all available models across modalities

API Reference

Complete API documentation