Fireworks AI Python SDK API library

The Fireworks AI Python SDK library provides convenient access to the Fireworks REST API from any Python 3.9+ application. The library includes type definitions for all request params and response fields, and offers both synchronous and asynchronous clients. The synchronous client uses httpx, while the asynchronous client uses aiohttp by default for improved concurrency performance.

Documentation

The REST API documentation can be found on docs.fireworks.ai. The full API of this library can be found in api.md.

Installation

# install from PyPI
pip install fireworks-ai

For using our Tinker-compatible API for training (GRPO/DPO), install the optional extras:

pip install 'fireworks-ai[training]'

Training SDK calls require a training-scoped Fireworks API key. Inference-only keys can still work for completions, but they return HTTP 401 on trainer and training-shape endpoints.

On Python 3.11+, the training extra installs the pinned tinker runtime used by the SDK. On Python 3.9 and 3.10, the base SDK still installs, but Tinker-backed training dependencies are intentionally omitted because the pinned Tinker runtime no longer supports those Python versions. Cookbook packages that use recipe helpers should declare tinker-cookbook directly so SDK-only environments do not install cookbook-only dependencies.

Note: The training SDK source lives entirely in src/fireworks/training/. All other code in this repository is auto-generated by Stainless and should not be referenced or modified when working on training features. See src/fireworks/training/README.md for training-specific documentation.

Usage

The full API of this library can be found in api.md.

import os
from fireworks import Fireworks

client = Fireworks(
    api_key=os.environ.get("FIREWORKS_API_KEY"),  # This is the default and can be omitted
)

completion = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "How do LLMs work?",
        }
    ],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
)
print(completion.choices[0].message.content)

While you can provide an api_key keyword argument, we recommend using python-dotenv to add FIREWORKS_API_KEY="My API Key" to your .env file so that your API Key is not stored in source control.

Async usage

Simply import AsyncFireworks instead of Fireworks and use await with each API call:

import os
import asyncio
from fireworks import AsyncFireworks

client = AsyncFireworks(
    api_key=os.environ.get("FIREWORKS_API_KEY"),  # This is the default and can be omitted
)


async def main() -> None:
    completion = await client.chat.completions.create(
        messages=[
            {
                "role": "user",
                "content": "How do LLMs work?",
            }
        ],
        model="accounts/fireworks/models/kimi-k2-instruct-0905",
    )
    print(completion.choices[0].message.content)


asyncio.run(main())

Functionality between the synchronous and asynchronous clients is otherwise identical.

Configuring account ID

Many Fireworks API methods require an account_id parameter (e.g., dataset operations, fine-tuning jobs, deployments). You can configure this at the client level to avoid passing it to every method call.

Setting account ID on the client

from fireworks import Fireworks

# Pass account_id during client initialization
client = Fireworks(
    account_id="my-account-id",
)

# Now you can omit account_id from method calls
dataset = client.datasets.create(
    dataset_id="my-dataset",
    dataset={"exampleCount": "100"},
)

# Instead of having to pass it every time:
# dataset = client.datasets.create(
#     account_id="my-account-id",  # No longer needed!
#     dataset_id="my-dataset",
#     dataset={"exampleCount": "100"},
# )

Using environment variables

You can also set the account ID using the FIREWORKS_ACCOUNT_ID environment variable:

export FIREWORKS_ACCOUNT_ID="my-account-id"

from fireworks import Fireworks

# account_id is automatically read from FIREWORKS_ACCOUNT_ID
client = Fireworks()

# All methods that need account_id will use the configured value
datasets = client.datasets.list()

Precedence

The account_id is resolved in the following order:

1. Explicitly passed to the method call (highest priority)
2. Set on the client during initialization
3. Read from FIREWORKS_ACCOUNT_ID environment variable

This means you can still override the client-level account_id for specific calls if needed:

client = Fireworks(account_id="default-account")

# Uses "default-account"
client.datasets.list()

# Override for this specific call
client.datasets.list(account_id="other-account")

Streaming responses

We provide support for streaming responses using Server Side Events (SSE).

from fireworks import Fireworks

client = Fireworks()

stream = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "How do LLMs work?",
        }
    ],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
    stream=True,
)
for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

The async client uses the exact same interface.

from fireworks import AsyncFireworks

client = AsyncFireworks()

stream = await client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "How do LLMs work?",
        }
    ],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
    stream=True,
)
async for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Reasoning models

For detailed information on how to use reasoning models with the Fireworks Python SDK, see the guide on reasoning.

Using types

Nested request parameters are TypedDicts. Responses are Pydantic models which also provide helper methods for things like:

• Serializing back into JSON, model.to_json()
• Converting to a dictionary, model.to_dict()

Typed requests and responses provide autocomplete and documentation within your editor. If you would like to see type errors in VS Code to help catch bugs earlier, set python.analysis.typeCheckingMode to basic.

Pagination

List methods in the Fireworks API are paginated.

This library provides auto-paginating iterators with each list response, so you do not have to request successive pages manually:

from fireworks import Fireworks

client = Fireworks()

all_batch_inference_jobs = []
# Automatically fetches more pages as needed.
for batch_inference_job in client.batch_inference_jobs.list(
    account_id="account_id",
):
    # Do something with batch_inference_job here
    all_batch_inference_jobs.append(batch_inference_job)
print(all_batch_inference_jobs)

Or, asynchronously:

import asyncio
from fireworks import AsyncFireworks

client = AsyncFireworks()


async def main() -> None:
    all_batch_inference_jobs = []
    # Iterate through items across all pages, issuing requests as needed.
    async for batch_inference_job in client.batch_inference_jobs.list(
        account_id="account_id",
    ):
        all_batch_inference_jobs.append(batch_inference_job)
    print(all_batch_inference_jobs)


asyncio.run(main())

Alternatively, you can use the .has_next_page(), .next_page_info(), or .get_next_page() methods for more granular control working with pages:

first_page = await client.batch_inference_jobs.list(
    account_id="account_id",
)
if first_page.has_next_page():
    print(f"will fetch next page using these details: {first_page.next_page_info()}")
    next_page = await first_page.get_next_page()
    print(f"number of items we just fetched: {len(next_page.batch_inference_jobs)}")

# Remove `await` for non-async usage.

Or just work directly with the returned data:

first_page = await client.batch_inference_jobs.list(
    account_id="account_id",
)

print(f"next page cursor: {first_page.next_page_token}")  # => "next page cursor: ..."
for batch_inference_job in first_page.batch_inference_jobs:
    print(batch_inference_job.continued_from_job_name)

# Remove `await` for non-async usage.

Nested params

Nested parameters are dictionaries, typed using TypedDict, for example:

from fireworks import Fireworks

client = Fireworks()

completion = client.chat.completions.create(
    messages=[{"role": "role"}],
    model="model",
    response_format={"type": "json_object"},
)
print(completion.response_format)

File uploads

Request parameters that correspond to file uploads can be passed as bytes, or a PathLike instance or a tuple of (filename, contents, media type).

from pathlib import Path
from fireworks import Fireworks

client = Fireworks()

client.datasets.upload(
    dataset_id="dataset_id",
    file=Path("/path/to/file"),
)

The async client uses the exact same interface. If you pass a PathLike instance, the file contents will be read asynchronously automatically.

Uploading datasets

The SDK provides two methods for uploading datasets to Fireworks, depending on file size.

Option 1: Direct upload (files < 150MB) - Recommended

For files under 150MB, use the streamlined direct upload approach. This uses only SDK methods with no additional dependencies:

import time
from pathlib import Path
from fireworks import Fireworks

client = Fireworks()

dataset_id = "my-dataset"
file_path = Path("/path/to/your-dataset.jsonl")

# Count lines in the dataset file (optional, for bookkeeping)
with open(file_path) as f:
    example_count = sum(1 for line in f if line.strip())

# Step 1: Create the dataset record
dataset = client.datasets.create(
    dataset_id=dataset_id,
    dataset={"exampleCount": str(example_count)},
)
print(f"Created dataset: {dataset.name}")

# Step 2: Upload the file
upload_response = client.datasets.upload(
    dataset_id=dataset_id,
    file=file_path,
)
print(f"Upload response: {upload_response}")

# Step 3: Poll until dataset is ready
while True:
    dataset = client.datasets.get(dataset_id=dataset_id)
    print(f"Dataset state: {dataset.state}")
    if dataset.state == "READY":
        print("Dataset is ready!")
        break
    elif dataset.state == "UPLOADING":
        time.sleep(2)
    else:
        raise Exception(f"Unexpected dataset state: {dataset.state}")

Option 2: Signed URL upload (files > 150MB)

For larger files, use the signed URL approach. This requires an HTTP client (like httpx or requests) to upload to the signed URL:

import time
from pathlib import Path
import httpx  # or use requests
from fireworks import Fireworks

client = Fireworks()

dataset_id = "my-large-dataset"
file_path = Path("/path/to/your-large-dataset.jsonl")
file_size = file_path.stat().st_size
file_name = file_path.name

# Count lines in the dataset file (optional, for bookkeeping)
with open(file_path) as f:
    example_count = sum(1 for line in f if line.strip())

# Step 1: Create the dataset record
dataset = client.datasets.create(
    dataset_id=dataset_id,
    dataset={"exampleCount": str(example_count)},
)
print(f"Created dataset: {dataset.name}")

# Step 2: Get signed upload URL
upload_endpoint = client.datasets.get_upload_endpoint(
    dataset_id=dataset_id,
    filename_to_size={file_name: str(file_size)},
)
signed_url = upload_endpoint.filename_to_signed_urls[file_name]
print(f"Got signed URL for upload")

# Step 3: Upload directly to the signed URL (requires external HTTP client)
with open(file_path, "rb") as f:
    file_content = f.read()

response = httpx.put(
    signed_url,
    content=file_content,
    headers={
        "Content-Type": "application/octet-stream",
        "x-goog-content-length-range": f"{file_size},{file_size}",
    },
)
response.raise_for_status()
print("File uploaded to signed URL")

# Step 4: Validate the upload
client.datasets.validate_upload(
    dataset_id=dataset_id,
    body={},
)
print("Upload validated")

# Step 5: Poll until dataset is ready
while True:
    dataset = client.datasets.get(dataset_id=dataset_id)
    print(f"Dataset state: {dataset.state}")
    if dataset.state == "READY":
        print("Dataset is ready!")
        break
    elif dataset.state == "UPLOADING":
        time.sleep(2)
    else:
        raise Exception(f"Unexpected dataset state: {dataset.state}")

Note: The httpx library is already a dependency of the SDK, so no additional installation is needed.

Dataset file format

Dataset files should be in JSONL format (JSON Lines), where each line is a valid JSON object. For chat-based fine-tuning, use the following format:

{"messages": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Hello!"}, {"role": "assistant", "content": "Hi there! How can I help you today?"}]}
{"messages": [{"role": "user", "content": "What is 2+2?"}, {"role": "assistant", "content": "2+2 equals 4."}]}

Managing datasets

You can view and manage your datasets in the Fireworks Dashboard.

# List all datasets
datasets = client.datasets.list()
for ds in datasets.datasets:
    print(f"{ds.name}: {ds.state}")

# Get a specific dataset
dataset = client.datasets.get(dataset_id=dataset_id)

# Delete a dataset
client.datasets.delete(dataset_id=dataset_id)

Handling errors

When the library is unable to connect to the API (for example, due to network connection problems or a timeout), a subclass of fireworks.APIConnectionError is raised.

When the API returns a non-success status code (that is, 4xx or 5xx response), a subclass of fireworks.APIStatusError is raised, containing status_code and response properties.

All errors inherit from fireworks.APIError.

import fireworks
from fireworks import Fireworks

client = Fireworks()

try:
    client.chat.completions.create(
        messages=[
            {
                "role": "user",
                "content": "How do LLMs work?",
            }
        ],
        model="accounts/fireworks/models/kimi-k2-instruct-0905",
    )
except fireworks.APIConnectionError as e:
    print("The server could not be reached")
    print(e.__cause__)  # an underlying Exception, likely raised within httpx.
except fireworks.RateLimitError as e:
    print("A 429 status code was received; we should back off a bit.")
except fireworks.APIStatusError as e:
    print("Another non-200-range status code was received")
    print(e.status_code)
    print(e.response)

Error codes are as follows:

Retries

Certain errors are automatically retried 2 times by default, with a short exponential backoff. Connection errors (for example, due to a network connectivity problem), 408 Request Timeout, 409 Conflict, 429 Rate Limit, and >=500 Internal errors are all retried by default.

You can use the max_retries option to configure or disable retry settings:

from fireworks import Fireworks

# Configure the default for all requests:
client = Fireworks(
    # default is 2
    max_retries=0,
)

# Or, configure per-request:
client.with_options(max_retries=5).chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "How do LLMs work?",
        }
    ],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
)

Timeouts

By default requests time out after 1 minute. You can configure this with a timeout option, which accepts a float or an httpx.Timeout object:

from fireworks import Fireworks

# Configure the default for all requests:
client = Fireworks(
    # 20 seconds (default is 1 minute)
    timeout=20.0,
)

# More granular control:
client = Fireworks(
    timeout=httpx.Timeout(60.0, read=5.0, write=10.0, connect=2.0),
)

# Override per-request:
client.with_options(timeout=5.0).chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "How do LLMs work?",
        }
    ],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
)

On timeout, an APITimeoutError is thrown.

Note that requests that time out are retried twice by default.

Advanced

Logging

We use the standard library logging module.

You can enable logging by setting the environment variable FIREWORKS_LOG to info.

$ export FIREWORKS_LOG=info

Or to debug for more verbose logging.

How to tell whether None means null or missing

In an API response, a field may be explicitly null, or missing entirely; in either case, its value is None in this library. You can differentiate the two cases with .model_fields_set:

if response.my_field is None:
  if 'my_field' not in response.model_fields_set:
    print('Got json like {}, without a "my_field" key present at all.')
  else:
    print('Got json like {"my_field": null}.')

Accessing raw response data (e.g. headers)

The "raw" Response object can be accessed by prefixing .with_raw_response. to any HTTP method call, e.g.,

from fireworks import Fireworks

client = Fireworks()
response = client.chat.completions.with_raw_response.create(
    messages=[{
        "role": "user",
        "content": "How do LLMs work?",
    }],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
)
print(response.headers.get('X-My-Header'))

completion = response.parse()  # get the object that `chat.completions.create()` would have returned
print(completion.id)

These methods return an APIResponse object.

The async client returns an AsyncAPIResponse with the same structure, the only difference being awaitable methods for reading the response content.

.with_streaming_response

The above interface eagerly reads the full response body when you make the request, which may not always be what you want.

To stream the response body, use .with_streaming_response instead, which requires a context manager and only reads the response body once you call .read(), .text(), .json(), .iter_bytes(), .iter_text(), .iter_lines() or .parse(). In the async client, these are async methods.

with client.chat.completions.with_streaming_response.create(
    messages=[
        {
            "role": "user",
            "content": "How do LLMs work?",
        }
    ],
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
) as response:
    print(response.headers.get("X-My-Header"))

    for line in response.iter_lines():
        print(line)

The context manager is required so that the response will reliably be closed.

Making custom/undocumented requests

This library is typed for convenient access to the documented API.

If you need to access undocumented endpoints, params, or response properties, the library can still be used.

Undocumented endpoints

To make requests to undocumented endpoints, you can make requests using client.get, client.post, and other http verbs. Options on the client will be respected (such as retries) when making this request.

import httpx

response = client.post(
    "/foo",
    cast_to=httpx.Response,
    body={"my_param": True},
)

print(response.headers.get("x-foo"))

Undocumented request params

If you want to explicitly send an extra param, you can do so with the extra_query, extra_body, and extra_headers request options.

Undocumented response properties

To access undocumented response properties, you can access the extra fields like response.unknown_prop. You can also get all the extra fields on the Pydantic model as a dict with response.model_extra.

Configuring the HTTP client

You can directly override the httpx client to customize it for your use case, including:

• Support for proxies
• Custom transports
• Additional advanced functionality

import httpx
from fireworks import Fireworks, DefaultHttpxClient

client = Fireworks(
    # Or use the `FIREWORKS_BASE_URL` env var
    base_url="http://my.test.server.example.com:8083",
    http_client=DefaultHttpxClient(
        proxy="http://my.test.proxy.example.com",
        transport=httpx.HTTPTransport(local_address="0.0.0.0"),
    ),
)

You can also customize the client on a per-request basis by using with_options():

client.with_options(http_client=DefaultHttpxClient(...))

Managing HTTP resources

By default the library closes underlying HTTP connections whenever the client is garbage collected. You can manually close the client using the .close() method if desired, or with a context manager that closes when exiting.

from fireworks import Fireworks

with Fireworks() as client:
  # make requests here
  ...

# HTTP client is now closed

Versioning

This package generally follows SemVer conventions, though certain backwards-incompatible changes may be released as minor versions:

1. Changes that only affect static types, without breaking runtime behavior.
2. Changes to library internals which are technically public but not intended or documented for external use. (Please open a GitHub issue to let us know if you are relying on such internals.)
3. Changes that we do not expect to impact the vast majority of users in practice.

We take backwards-compatibility seriously and work hard to ensure you can rely on a smooth upgrade experience.

We are keen for your feedback; please open an issue with questions, bugs, or suggestions.

Determining the installed version

If you've upgraded to the latest version but aren't seeing any new features you were expecting then your python environment is likely still using an older version.

You can determine the version that is being used at runtime with:

import fireworks
print(fireworks.__version__)

Requirements

Python 3.9 or higher.

Contributing

See the contributing documentation.