Fitantanana hadisoana

Ny SDK dia manangana object hadisoana mitondra ny code status sy ny saha ao amin'ny body. Ity fangatahana ity dia manonona model tsy misy:

import openai
from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.shannon-ai.com/v1",
)

try:
    client.chat.completions.create(
        model="no-such-model",
        messages=[{"role": "user", "content": "Hello"}],
    )
except openai.APIStatusError as error:
    print(error.status_code)                           # 400
    print(error.type)                                  # invalid_request_error
    print(error.body["message"])                       # unknown model: no-such-model
    print(error.response.headers.get("x-request-id"))  # the id of this request

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://api.shannon-ai.com/v1",
});

try {
  await client.chat.completions.create({
    model: "no-such-model",
    messages: [{ role: "user", content: "Hello" }],
  });
} catch (error) {
  if (!(error instanceof OpenAI.APIError)) throw error;
  console.log(error.status);         // 400
  console.log(error.type);           // invalid_request_error
  console.log(error.error.message);  // unknown model: no-such-model
}

curl -i https://api.shannon-ai.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "no-such-model",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Mamerina status 400 izy ary ity body ity:

{
  "error": {
    "type": "invalid_request_error",
    "message": "unknown model: no-such-model"
  }
}

Vatan'ny hadisoana

Misy body roa. Ny /v1/messages dia mamaly amin'ny endrika Anthropic; ny endpoint hafa rehetra dia mamaly amin'ny endrika OpenAI. Amin'ireo roa ireo, ny antsipiriany dia ao amin'ny error.type sy error.message, ka mpamaky iray no ampy ho an'ny endpoint rehetra.

{
  "error": {
    "type": "authentication_error",
    "message": "Invalid API key",
    "code": "invalid_api_key",
    "param": null
  }
}

{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "Invalid API key"
  }
}

Karazan'ny hadisoana

Code status

Atao araka ity filaharana ity ny fanamarinana: API key, body ny fangatahana, id ny model, flood protection, balance. Noho izany, ny fangatahana misy key diso sy body diso dia mamerina 401. Ny fangatahana nolavina tamin'ny status 4xx dia tsy vidiana.

Balance sy flood protection

Ny fangatahana rehetra dia tanterahina mitovy: tsy misy rate tier ary tsy misy quota API mitokana. Misy zavatra roa mbola mety mamaly 429, samy manana ny type rate_limit_error. Ny hafatra no manavaka azy.

Balance

Ny balance-nao dia ny allowance an'ny plan androany, izay averina amin'ny 00:00 UTC, ary raha lany dia ny credit novidinao. Ny chat sy ny API dia mamoaka avy amin'ny balance iray ihany.

Ny fangatahana tsirairay dia mametraka voalohany ny fetran'ny output-ny: max_tokens, na max_output_tokens amin'ny /v1/responses, 4,096 raha tsy mandefa roa ianao. Rehefa vita ny valiny, ny habetsaky ny napetraka dia soloina ny nampiasain'ny fangatahana. Raha tsy mifanaraka amin'ny balance sisa ny habetsaka, ny fangatahana dia mamerina Quota exceeded alohan'ny hanombohan'ny model.

Ny fangatahana mandeha miaraka dia samy mametraka ny habetsahany. Noho izany, fangatahana maro miaraka misy max_tokens avo dia mety hampitsahatra vetivety ny balance, na dia fohy aza ny valiny tsirairay. Ny max_tokens ambany kokoa dia mametraka kely kokoa.

Flood protection

Ny flood protection dia mamela fangatahana 120 isa-minitra isaky ny kaonty. Ny kaonty no ianarana, fa tsy ny key, amin'ny /v1/chat/completions, /v1/responses ary /v1/messages miaraka. Manomboka amin'ny fangatahana voalohany ny minitra.

Ny fangatahana faha-121 ao anatin'io minitra io dia mamerina Too many requests. Retry in <n>s. ary tsy misy vidiny. Ianarana koa ny fangatahana nolavina, ka ny client izay mandefa indray tsy miandry dia mijanona nolavina mandra-pahatapitry ny minitra. Andraso ny <n> segondra voalaza ao amin'ny hafatra.

Ny fangatahana alefa miaraka dia tsy lavina noho izy miaraka. Miandry ao anaty filaharana izy ireo ary valiana isam-bava.

Ny GET /v1/models sy ny endpoint fanisana token dia tsy ianarana.

Fanampiana bebe kokoa momba ny fetra sy ny balance, ary izay hijerena ny balance sy ny fampiasanao: Fetra sy balance Keys & usage

Izay hatao retry

Alefaso indray ny fangatahana rehefa mety hahomby ny fangatahana mitovy ihany. Ny SDK OpenAI sy Anthropic dia efa manao retry indroa ho azy amin'ny valiny 429 sy 5xx; apetraho ny max_retries (maxRetries amin'ny JavaScript) raha te-hanova izany.

Retry loop manaraka ity tabilao ity:

import re
import time

import openai
from openai import OpenAI

# max_retries=0: this loop does the retrying, the SDK does not add its own.
client = OpenAI(api_key="YOUR_API_KEY", base_url="https://api.shannon-ai.com/v1", max_retries=0)


def create_with_retry(attempts=5, **request):
    for attempt in range(attempts):
        try:
            return client.chat.completions.create(**request)
        except openai.RateLimitError as error:
            text = str(error)
            flood = re.search(r"Retry in (\d+)s", text)
            if flood:
                time.sleep(int(flood.group(1)))   # flood protection: wait out the minute
            elif "temporarily busy" in text:
                time.sleep(2 ** attempt)
            else:
                raise                             # balance: a wait of seconds does not help
        except (openai.InternalServerError, openai.APIConnectionError):
            time.sleep(2 ** attempt)              # 1, 2, 4, 8 seconds
    raise RuntimeError("the request failed after all attempts")


response = create_with_retry(
    model="shannon-3",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

import OpenAI from "openai";

// maxRetries: 0 — this loop does the retrying, the SDK does not add its own.
const client = new OpenAI({ apiKey: "YOUR_API_KEY", baseURL: "https://api.shannon-ai.com/v1", maxRetries: 0 });

const sleep = (seconds) => new Promise((resolve) => setTimeout(resolve, seconds * 1000));

async function createWithRetry(request, attempts = 5) {
  for (let attempt = 0; attempt < attempts; attempt++) {
    try {
      return await client.chat.completions.create(request);
    } catch (error) {
      const text = String(error.message);
      const flood = text.match(/Retry in (\d+)s/);
      if (error.status === 429 && flood) {
        await sleep(Number(flood[1]));            // flood protection: wait out the minute
      } else if (error.status === 429 && text.includes("temporarily busy")) {
        await sleep(2 ** attempt);
      } else if (error.status >= 500 || error instanceof OpenAI.APIConnectionError) {
        await sleep(2 ** attempt);                // 1, 2, 4, 8 seconds
      } else {
        throw error;                              // wrong request, wrong key or balance
      }
    }
  }
  throw new Error("the request failed after all attempts");
}

const response = await createWithRetry({
  model: "shannon-3",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(response.choices[0].message.content);

Hadisoana ao anaty stream

Ny stream dia manana status 200 hatrany amin'ny byte voalohany. Ny fahatafintohinana aorian'izay dia tonga ho event ao anaty stream: chunk misy mpikambana error amin'ny /v1/chat/completions, event error amin'ny /v1/messages, event response.failed na error amin'ny /v1/responses. Mbola manaraka ireo event famaranana ny stream. Heverina ho tsy nahomby ny valiny toy izany ary alefaso indray ny fangatahana, toy ny amin'ny status 5xx.

Ny resaka lava noho ny context window an'ny model dia ambara amin'izany fomba izany, miaraka amin'ny type invalid_request_error sy ny code context_length_exceeded. Ahenoy ny resaka alohan'ny handefasanao indray.

Ny frame marina dia voasoratra miaraka amin'ny streaming. Fandefasana mivantana

Id an'ny fangatahana

Ny valiny rehetra, anisan'izany ny hadisoana, dia manana header x-request-id. Io dia ny sanda nalefanao tao amin'ny header x-request-id anao, na tarehintsoratra hexadecimal 12 raha tsy nandefa ianao. Tehirizo ao amin'ny log izy, ary lazao rehefa mangataka fanampiana momba ny fangatahana.