Text Generation Inference (TGI) Alternative: Integrating with Fikra API

Overview

Text Generation Inference (TGI) is an external inference framework. Fikra API does not host TGI or claim compatibility with TGI-specific protocols. However, Fikra API provides an OpenAI-style chat-completions interface. You can use standard OpenAI-compatible SDKs or raw HTTP clients to interact with Fikra API, effectively replacing a local TGI deployment with a managed API endpoint.

Prerequisites

1. A Fikra API account with a valid API key (fk_live_...).
2. Python 3.8+ or Node.js 16+.
3. The openai Python package or openai Node.js package (or any HTTP client).

Environment Variable Setup

Store your API key in an environment variable to avoid hardcoding secrets.

export FIKRA_API_KEY="fk_live_your_key_here"

Python Example

The following example uses the openai Python package configured to point to the Fikra API base URL.

import os
from openai import OpenAI

# Initialize client with Fikra API base URL
client = OpenAI(
    base_url="https://api.fikraapi.co.ke/v1",
    api_key=os.getenv("FIKRA_API_KEY")
)

response = client.chat.completions.create(
    model="fikra-flash",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello, Fikra!"}
    ]
)

print(response.choices[0].message.content)

Streaming Guidance

Fikra API supports streaming via Server-Sent Events (SSE) when stream: true is set. The response will be a stream of chunks.

stream = client.chat.completions.create(
    model="fikra-flash",
    messages=[
        {"role": "user", "content": "Explain quantum computing in simple terms."}
    ],
    stream=True
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Common Errors

• 400 Bad Request: Malformed JSON, missing required fields, or invalid model alias. Check the model registry for available aliases.
• 401 Unauthorized: Missing or invalid Authorization header. Ensure the Bearer token is correctly formatted.
• 402 Payment Required: Insufficient token credits in your wallet. Top up via the Fikra dashboard.
• 429 Too Many Requests: Rate limit exceeded. Unverified accounts are limited to 30 RPM; trusted accounts to 100 RPM. Implement exponential backoff.
• 503 Service Unavailable: Temporary server issue. Retry with backoff.

Production Checklist

1. API Key Management: Use environment variables or a secrets manager. Never commit keys to version control.
2. Rate Limiting: Implement client-side throttling to stay within 30 RPM (unverified) or 100 RPM (trusted) limits.
3. Error Handling: Handle 429 and 503 errors with retry logic and exponential backoff.
4. Monitoring: Log request latency and error rates. Monitor token credit usage via the Fikra dashboard.
5. Model Selection: Use fikra-flash for low-latency tasks. Check the model registry for other aliases like fikra-pro-20b if higher quality is needed.
6. Security: Use HTTPS for all requests. Validate input to prevent prompt injection if applicable to your use case.