Legacy text completions

Use prompt-based completions with JSON and SSE responses.

POST /v1/completions accepts a text prompt, not a messages conversation. Confirm endpoint support with the deployment operator; new applications generally start with Chat or Responses. An SDK method does not prove a chat model supports this endpoint.

This example requests one text completion and reads choices[0].text, requiring finish_reason=stop. Streaming increments also use text, not Chat’s delta.content. Length limits, filtering, truncated streams, and empty text fail validation. Output is buffered until completion. The SDK parses SSE; the example validates completion fields and provides no stream resumption.

Install dependencies and set API_BASE_URL (including /v1), API_KEY, and API_MODEL as in Python SDK.

.venv/bin/python examples/legacy-completions.py sync
.venv/bin/python examples/legacy-completions.py stream

See the API reference. Preservation of legacy fields such as suffix, echo, logprobs, or best_of depends on the converter. Protocol reference: OpenAI Completions; gateway behavior follows this project’s implementation.

import os
import sys
from openai import OpenAI, APIConnectionError, APIStatusError

mode = sys.argv[1] if len(sys.argv) == 2 else "sync"
if mode not in ("sync", "stream"):
    raise SystemExit("Usage: python legacy-completions.py [sync|stream]")
with OpenAI(base_url=os.environ["API_BASE_URL"].rstrip("/"),
            api_key=os.environ["API_KEY"], timeout=60.0, max_retries=0) as client:
    try:
        result = client.completions.create(model=os.environ["API_MODEL"],
            prompt="Complete this sentence: A good API guide should", stream=mode == "stream")
        text, finished = "", False
        if mode == "sync":
            if len(result.choices) != 1 or result.choices[0].index != 0:
                raise ValueError("Expected one completion.")
            choice = result.choices[0]
            if choice.finish_reason != "stop" or not isinstance(choice.text, str):
                raise ValueError("Incomplete completion.")
            text, finished = choice.text, True
        else:
            with result:
                for chunk in result:
                    if not chunk.choices:
                        continue  # Usage-only chunk, not a completion marker.
                    if finished or len(chunk.choices) != 1 or chunk.choices[0].index != 0:
                        raise ValueError("Unexpected completion chunk.")
                    choice = chunk.choices[0]
                    if not isinstance(choice.text, str) or choice.finish_reason not in (None, "stop"):
                        raise ValueError("Incomplete completion.")
                    text += choice.text
                    if len(text) > 1_000_000:
                        raise ValueError("Example output limit exceeded.")
                    finished = choice.finish_reason == "stop"
        if not finished or not text.strip():
            raise ValueError("Missing completed text.")
        print(text)
    except APIStatusError as error:
        raise SystemExit(f"Request failed (HTTP {error.status_code}). No automatic retry.")
    except APIConnectionError:
        raise SystemExit("Connection failed. No automatic retry.")
    except Exception:
        raise SystemExit("Completion did not produce valid complete text.")

阅读中文版