Fireworks.ai 为其 LLM 支持函数调用，与 OpenAI 类似。这让用户可以直接描述可用的工具/函数集合，并让模型动态选择要调用的正确函数，而无需用户进行复杂的提示。

由于我们的 Fireworks LLM 直接继承自 OpenAI，我们可以将现有抽象与 Fireworks 一起使用。

我们从三个层面展示这一点：直接在模型 API 上、作为 Pydantic Program（结构化输出提取）的一部分，以及作为智能体的一部分。

%pip install llama-index-llms-fireworks

%pip install llama-index

import os

os.environ["FIREWORKS_API_KEY"] = "fw_3ZkvBpQyjRzbicpihhrihaEP"

from llama_index.llms.fireworks import Fireworks

## define fireworks model, for a list of function calling models see: https://app.fireworks.ai/models/?filter=LLM&functionCalling=true
llm = Fireworks(
    model="accounts/fireworks/models/deepseek-v3p1-terminus", temperature=0
)

LLM 模块上的函数调用

你可以直接在 LLM 模块上输入函数调用。

import os
import json
from openai import OpenAI
from pydantic import BaseModel, Field
from llama_index.llms.openai.utils import to_openai_tool

class Song(BaseModel):
    """A song with name and artist"""

    name: str = Field(description="The name of the song")
    artist: str = Field(description="The artist who performed the song")

song_fn = to_openai_tool(Song)

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

response = client.chat.completions.create(
    model="accounts/fireworks/models/kimi-k2-instruct-0905",
    messages=[{"role": "user", "content": "Generate a song from Beyonce"}],
    tools=[song_fn],
    temperature=0.1,
)

print(response)

if response.choices[0].message.tool_calls:
    tool_call = response.choices[0].message.tool_calls[0]
    print(f"\nTool called: {tool_call.function.name}")

    # Parse the arguments to get structured output
    arguments = json.loads(tool_call.function.arguments)
    print(f"Arguments: {arguments}")

    # Create Song instance from the structured output
    song = Song(**arguments)
    print(f"\nExtracted Song:")
    print(f"Name: {song.name}")
    print(f"Artist: {song.artist}")

ChatCompletion(id='07921e74-5dca-409c-a4d3-1a2e0c7cd1e7', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content='```json\n{\n  "name": "Halo",\n  "artist": "Beyoncé"\n}\n```', refusal=None, role='assistant', annotations=None, audio=None, function_call=None, tool_calls=None))], created=1761704700, model='accounts/fireworks/models/kimi-k2-instruct-0905', object='chat.completion', service_tier=None, system_fingerprint=None, usage=CompletionUsage(completion_tokens=25, prompt_tokens=145, total_tokens=170, completion_tokens_details=None, prompt_tokens_details=PromptTokensDetails(audio_tokens=None, cached_tokens=0)))