Inception

總覽

屬性	詳細資料
說明	Inception 提供 Mercury 系列的 diffusion LLM（dLLMs）。此 API 與 OpenAI 相容。
LiteLLM 上的提供者路由	inception/（chat）、text-completion-inception/（fill-in-the-middle）
提供者文件連結	Inception Platform Documentation ↗
基底 URL	https://api.inceptionlabs.ai/v1
支援的操作	/chat/completions、/fim/completions

可用模型

模型	說明	上下文視窗
inception/mercury-2	快速推理聊天模型；支援工具呼叫與結構化輸出	128,000 tokens
text-completion-inception/mercury-edit-2	用於 fill-in-the-middle（FIM）自動完成的程式碼模型	32,000 tokens

必要變數

Environment Variables

os.environ["INCEPTION_API_KEY"] = ""  # your Inception API key

使用方式 - LiteLLM Python SDK

非串流

Inception Non-streaming Completion

import os
import litellm
from litellm import completion

os.environ["INCEPTION_API_KEY"] = ""  # your Inception API key

messages = [{"content": "Hello, how are you?", "role": "user"}]

# Inception call
response = completion(
    model="inception/mercury-2",
    messages=messages
)

print(response)

串流

Inception Streaming Completion

import os
import litellm
from litellm import completion

os.environ["INCEPTION_API_KEY"] = ""  # your Inception API key

messages = [{"content": "Write a short story about AI", "role": "user"}]

# Inception call with streaming
response = completion(
    model="inception/mercury-2",
    messages=messages,
    stream=True
)

for chunk in response:
    print(chunk)

推理努力與推理摘要

Mercury 提供一個 reasoning_effort 控制項，並具備 Inception 專屬的 instant 值，可用於近即時回應；同時也提供標準的 low、medium 與 high。設定 reasoning_summary=True 以接收模型在回應中的推理摘要。

Inception Reasoning

import os
from litellm import completion

os.environ["INCEPTION_API_KEY"] = ""  # your Inception API key

response = completion(
    model="inception/mercury-2",
    messages=[{"role": "user", "content": "If a bat and ball cost $1.10 and the bat is $1 more than the ball, how much is the ball?"}],
    reasoning_effort="high",
    reasoning_summary=True,
)

print(response.choices[0].message.content)
print(response.reasoning_summary)  # {"content": "...", "status": "complete"}

函式呼叫

Inception Function Calling

import os
from litellm import completion

os.environ["INCEPTION_API_KEY"] = ""  # your Inception API key

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather in a location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "The city and state, e.g. San Francisco, CA"
                }
            },
            "required": ["location"]
        }
    }
}]

messages = [{"role": "user", "content": "What's the weather in Boston?"}]

response = completion(
    model="inception/mercury-2",
    messages=messages,
    tools=tools,
    tool_choice="auto"
)

print(response)

中間填補（FIM）

mercury-edit-2 透過 Inception 的 /v1/fim/completions 端點提供程式碼自動完成。搭配 text_completion 與 text-completion-inception/ 路由使用，並傳入一個 prompt（前綴）以及可選的 suffix。

Inception FIM

import os
from litellm import text_completion

os.environ["INCEPTION_API_KEY"] = ""  # your Inception API key

response = text_completion(
    model="text-completion-inception/mercury-edit-2",
    prompt="def add(a, b):\n    return ",
    suffix="\n",
    max_tokens=64,
)

print(response.choices[0].text)

使用方式 - LiteLLM Proxy Server

config.yaml

model_list:
  - model_name: mercury-2
    litellm_params:
      model: inception/mercury-2
      api_key: os.environ/INCEPTION_API_KEY
  - model_name: mercury-edit-2
    litellm_params:
      model: text-completion-inception/mercury-edit-2
      api_key: os.environ/INCEPTION_API_KEY

支援的 OpenAI 參數

• max_tokens
• max_completion_tokens
• temperature
• stop
• tools
• tool_choice
• stream
• stream_options
• response_format

Inception 專屬參數

這些參數會轉送到 Inception chat API：

• reasoning_effort (instant | low | medium | high)
• reasoning_summary（bool）— 傳回模型推理摘要
• reasoning_summary_wait（bool）— 在傳回前等待摘要完成
• diffusing（bool）— 串流中間去噪步驟
• realtime（bool）— 針對最低延遲進行最佳化