Advanced Configuration

English | 中文 | 日本語 | Español | 한국어

This guide covers SDK-wide configuration beyond authentication and region selection. For API key setup, see API Key Authentication; for switching regions/endpoints, see Region and Endpoint Configuration.

Request Timeout

Every .call()/.create()/.get()/.list()/.delete()-style method accepts a request_timeout keyword argument (in seconds). It defaults to 300 seconds (DEFAULT_REQUEST_TIMEOUT_SECONDS). For streaming requests this is the idle timeout between chunks; for non-streaming requests it's the total request timeout.

from dashscope import Generation

response = Generation.call(
    model="qwen-plus",
    messages=[{"role": "user", "content": "Hi"}],
    request_timeout=30,
)

Custom Request Headers

Pass a headers dict to merge additional HTTP headers into the request (the SDK's own Authorization/workspace headers are still applied):

from dashscope import Generation

response = Generation.call(
    model="qwen-plus",
    messages=[{"role": "user", "content": "Hi"}],
    headers={"X-Request-Tag": "my-service"},
)

The OpenAI-compatible Completions.create has a dedicated extra_headers parameter for the same purpose:

from dashscope.aigc.chat_completion import Completions

response = Completions.create(
    model="qwen-max",
    messages=[{"role": "user", "content": "Hi"}],
    api_key="YOUR-DASHSCOPE-API-KEY",
    extra_headers={"X-Request-Tag": "my-service"},
)

Proxy Support

The SDK doesn't implement its own proxy configuration; instead it relies on the underlying HTTP libraries' standard behavior. The async client creates its aiohttp.ClientSession with trust_env=True, and the sync client's requests.Session honors proxies by default — both automatically pick up the standard HTTP_PROXY, HTTPS_PROXY, and NO_PROXY environment variables:

export HTTPS_PROXY="http://proxy.example.com:8080"

from dashscope import Generation

# Requests now go through the proxy configured via HTTPS_PROXY
response = Generation.call(model="qwen-plus", messages=[{"role": "user", "content": "Hi"}])

Closing Shared Connection Pools

The SDK maintains a shared requests.Session (sync) and one aiohttp.ClientSession per event loop (async) for connection pooling. Long-running processes can close them explicitly to release pooled connections (a fresh session is created lazily on the next request):

import asyncio
from dashscope import close_shared_sync_session, close_shared_aio_session

close_shared_sync_session()
asyncio.run(close_shared_aio_session())

Logging

As covered in the README, setting DASHSCOPE_LOGGING_LEVEL before import attaches a console handler automatically. Since the SDK logs through the standard logging module under the "dashscope" logger name, you can also control it programmatically without the environment variable:

import logging

logging.getLogger("dashscope").setLevel(logging.DEBUG)

FilesExpand file tree

configuration.md

Latest commit

History