OpenAI SDK
官方 OpenAI SDK 无需改动即可使用。把 Base URL 设为 https://api.abliteratedllm.dev/v1,并使用你的 API 密钥。chat completions 和无状态的 Responses API 都可以使用。
占位符:
$API_BASE为https://api.abliteratedllm.dev/v1,$API_ORIGIN为https://api.abliteratedllm.dev,$YOUR_KEY为你的 API 密钥。
两个 SDK 也会读取 OPENAI_BASE_URL 和 OPENAI_API_KEY 环境变量,因此现有代码无需修改即可切换:
export OPENAI_BASE_URL="$API_BASE"
export OPENAI_API_KEY="$YOUR_KEY"Python
pip install openaiChat completions(流式输出):
from openai import OpenAI
client = OpenAI(base_url="$API_BASE", api_key="$YOUR_KEY")
stream = client.chat.completions.create(
model="glm-5.3-flash-abliterated",
messages=[
{"role": "system", "content": "You are a witty fantasy narrator."},
{"role": "user", "content": "Describe a tavern brawl in four sentences."},
],
reasoning_effort="low",
max_tokens=1024,
stream=True,
)
for chunk in stream:
# Usage arrives in a final chunk whose choices list is empty.
if chunk.choices:
print(chunk.choices[0].delta.content or "", end="", flush=True)
if chunk.usage:
print(f"\n\n{chunk.usage.prompt_tokens} in, {chunk.usage.completion_tokens} out")Responses API(无状态,因此每次都要发送完整对话):
response = client.responses.create(
model="glm-5.3-flash-abliterated",
instructions="Answer in one sentence.",
input="What is the boiling point of water at sea level in Fahrenheit?",
reasoning={"effort": "low"},
)
print(response.output_text)JavaScript 与 TypeScript
npm install openaiChat completions(流式输出):
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "$API_BASE", apiKey: "$YOUR_KEY" });
const stream = await client.chat.completions.create({
model: "glm-5.3-flash-abliterated",
messages: [
{ role: "system", content: "You are a witty fantasy narrator." },
{ role: "user", content: "Describe a tavern brawl in four sentences." },
],
reasoning_effort: "low",
max_tokens: 1024,
stream: true,
});
for await (const chunk of stream) {
// Usage arrives in a final chunk whose choices array is empty.
process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
if (chunk.usage) console.log(`\n\n${chunk.usage.prompt_tokens} in, ${chunk.usage.completion_tokens} out`);
}Responses API:
const response = await client.responses.create({
model: "glm-5.3-flash-abliterated",
instructions: "Answer in one sentence.",
input: "What is the boiling point of water at sea level in Fahrenheit?",
reasoning: { effort: "low" },
});
console.log(response.output_text);注意事项
- 错误采用 OpenAI 格式,因此 SDK 会抛出它们常规的异常(
RateLimitError、AuthenticationError等)。每个错误都会在code中带有我们的某个错误码。例外情况是运行超过 90 秒仍未结束的非流式请求:此时状态码已经是 200,因此之后发生的错误会出现在响应体中。检查顶层是否有error成员;参见长时间请求。 - 重试:默认情况下,SDK 会对 429 和 5xx 响应重试几次,并遵循我们随
capacity_exhausted和 503model_unavailable(该模型当前没有可用的服务容量)一起返回的Retry-After响应头。model_unavailable导致的服务中断可能持续几分钟,因此如果重试次数用完,就切换到其他模型。参见错误。 - Responses 是无状态的:
store始终为false,previous_response_id会返回 400stateless_only。参见 POST /v1/responses。 - 图像和视频:图像输入适用于 GET /v1/models 中
input_modalities包含image的模型,video_url(仅限 https URL)适用于包含video的模型。其他模型会以 400unsupported_content拒绝这些输入。音频和文件输入在所有模型上都会以同样的方式失败。参见 POST /v1/chat/completions。
已于 2026 年 11 月 3 日对照 OpenAI SDK 文档核实:openai-python(base_url、OPENAI_BASE_URL)和 openai-node(baseURL、OPENAI_BASE_URL)。