OpenAI compatibility

The LLM API is OpenAI compatible!

OpenAI Python client library

You can use OpenAI Python client library to interact with Fireworks. This makes migration of existing applications already using OpenAI particularly easy.

Specify endpoint and API key

You can override parameters for the entire application using environment variables


or by setting these values in Python

import openai

# warning: it has a process-wide effect
openai.api_base = ""
openai.api_key = "<YOUR_FIREWORKS_API_KEY>"

Alternatively, you may specify these parameters for a single request (useful if you mix calls to OpenAI and Fireworks in the same process):

# api_base and api_key can be passed to any of the supported APIs
chat_completion = openai.ChatCompletion.create(
    api_key = "<YOUR_FIREWORKS_API_KEY",


Use OpenAI's SDK how you'd normally would. Just ensure that the model parameter refers to one of Fireworks models.


Simple completion API that doesn't modify provided prompt in any way

completion = openai.Completion.create(
  prompt="The quick brown fox",

Chat Completion

Works best for models fine-tuned for conversation (e.g. llama*-chat variants)

chat_completion = openai.ChatCompletion.create(
      "role": "system",
     	"content": "You are a helpful assistant.",
      "role": "user",
      "content": "Say this is a test",

API compatibility


The following options have minor differences:

  • stop: the returned string includes the stop word for Fireworks while it's omitted for OpenAI (it can be easily truncated on client side)
  • max_tokens: behaves differently if the model context length is exceeded. If the length of prompt or messages plus max_tokens is higher than the model's context window, max_tokens will be adjusted lower accordingly. OpenAI returns invalid request error in this situation. This behavior can be adjusted by context_length_exceeded_behavior parameter.

Token usage for streaming responses

OpenAI API returns usage stats (number of tokens in prompt and completion) for non-streaming responses but doesn't for the streaming ones (see forum post). returns usage stats in both cases. For streaming responses, the usage field is returned in the very last chunk on the response (i.e. the one having finish_reason set). For example:

curl --request POST \           
     --url \
     --header "accept: application/json" \
     --header "authorization: Bearer $API_KEY" \
     --header "content-type: application/json" \
     --data '{"model": "accounts/fireworks/models/starcoder-16b-w8a16", "prompt": "def say_hello_world():", "max_tokens": 100, "stream": true}'

data: {..., "choices":[{"text":"\n  print('Hello,","index":0,"finish_reason":null,"logprobs":null}],"usage":null}
data: {..., "choices":[{"text":" World!')\n\n\n","index":0,"finish_reason":null,"logprobs":null}],"usage":null}
data: {..., "choices":[{"text":"say_hello_","index":0,"finish_reason":null,"logprobs":null}],"usage":null}
data: {..., "choices":[{"text":"world()\n","index":0,"finish_reason":"stop","logprobs":null}],"usage":{"prompt_tokens":7,"total_tokens":24,"completion_tokens":17}}

data: [DONE]

Note, that if you're using OpenAI SDK, they usage field won't be listed in the SDK's structure definition. But it can be accessed directly. For example:

  • In Python SDK, you can access the attribute directly, e.g. for chunk in openai.ChatCompletion.create(...): print(chunk["usage"]).
  • In TypeScript SDK, you need to cast away the typing, e.g. for await (const chunk of await { console.log((chunk as any).usage); }.

Not supported options

The following options are not yet supported:

  • presence_penalty
  • frequency_penalty
  • best_of: you can use n instead
  • logit_bias
  • functions: you can use our LangChain integration to achieve similar functionality client-side

Please reach out to us on Discord if you have a use case requiring one of these.