Troubleshooting DeepSeek Capacity Reached Errors

Understanding the Error

If you encounter a "DeepSeek capacity reached" error, note that Fikra API does not serve DeepSeek models. DeepSeek is an external provider. Fikra API offers its own model aliases (e.g., fikra-flash, fikra-pro-20b). If you are attempting to use a DeepSeek model alias in Fikra, the request will fail because the model is not supported.

Common HTTP Status Codes

When integrating with Fikra API, handle the following standard status codes:

• 400 Bad Request: The request is malformed or references an unsupported model alias. Verify that you are using a valid Fikra model alias from the registry.
• 401 Unauthorized: The API key is missing, invalid, or inactive. Ensure your Authorization: Bearer fk_live_... header is correctly formatted and the key is active in your Fikra dashboard.
• 402 Payment Required: Your token-credit wallet balance is insufficient. Top up your wallet via the Fikra dashboard using Paystack (M-Pesa or cards).
• 429 Too Many Requests: You have exceeded the rate limit. Unverified accounts are limited to 30 RPM; trusted accounts to 100 RPM.
• 503 Service Unavailable: The upstream provider is temporarily unavailable. This may occur if the underlying infrastructure for a Fikra model is experiencing transient issues.

Handling 429 and 503 with Exponential Backoff

For 429 and 503 errors, implement exponential backoff to retry requests safely. Do not assume automatic failover or dedicated capacity.

import time
import random
import requests

base_url = "https://api.fikraapi.co.ke/v1"
api_key = "YOUR_FIKRA_API_KEY"  # Use environment variable in production

headers = {
    "Authorization": f"Bearer {api_key}",
    "Content-Type": "application/json"
}

payload = {
    "model": "fikra-flash",
    "messages": [{"role": "user", "content": "Hello"}]
}

def request_with_backoff(max_retries=5):
    for attempt in range(max_retries):
        response = requests.post(f"{base_url}/chat/completions", headers=headers, json=payload)

        if response.status_code in [429, 503]:
            # Exponential backoff with jitter
            wait_time = (2 ** attempt) + random.uniform(0, 1)
            print(f"Received {response.status_code}. Retrying in {wait_time:.2f}s...")
            time.sleep(wait_time)
            continue

        if response.status_code == 200:
            return response.json()
        else:
            # Handle other errors (400, 401, 402) immediately
            raise Exception(f"Request failed with status {response.status_code}: {response.text}")

    raise Exception("Max retries exceeded for 429/503 errors.")

# Example usage
# result = request_with_backoff()

Recommendations

1. Verify Model Aliases: Ensure you are using only Fikra-supported model aliases (e.g., fikra-flash, fikra-pro-20b). Do not use external provider model names.
2. Monitor Rate Limits: Track your request rate to stay within 30 RPM (unverified) or 100 RPM (trusted) limits.
3. Check Wallet Balance: Ensure your token-credit wallet has sufficient balance to avoid 402 errors.
4. Retry Transient Errors: Use exponential backoff for 429 and 503 errors, but do not assume the issue will resolve without checking the Fikra dashboard for status updates.