Codestral API Tutorial: Getting Started With Mistral's API

Codestral: A Comprehensive Guide to the Code Generation API

Codestral, a cutting-edge generative model, excels at code generation tasks like fill-in-the-middle (FIM) and code completion. Trained on over 80 programming languages, it's a versatile tool for developers working with both common and less-used languages. This tutorial details how to effectively utilize the Codestral API. For a broader overview of Codestral, see my article on "What is Mistral's Codestral".

API Endpoints

Codestral offers two primary API endpoints:

• codestral.mistral.ai: Ideal for individual users and small-scale projects. Currently free (until August 1, 2024), it will transition to a subscription model.
• api.mistral.ai: Designed for business needs and high-volume usage, offering increased rate limits and robust support.

Mistral recommends codestral.mistral.ai for IDE plugins or user-facing tools, allowing users to manage their own API keys. api.mistral.ai is preferred for other applications due to its higher rate limits and scalability. This tutorial focuses on codestral.mistral.ai.

Getting Started

Obtaining an API Key:

1. Sign up: Create a Mistral AI account.
2. Get your API key: For api.mistral.ai, navigate to the API Keys tab and generate a new key. For codestral.mistral.ai, go to the Codestral tab (often marked "New"), complete the signup (note: a phone number is usually required), and access your key once approved.

Authentication (Python):

We'll use the requests library to create authentication functions for both endpoints:

import requests
import json

api_key = 'INSERT YOUR API KEY HERE'

def call_chat_endpoint(data, api_key=api_key):
    url = "https://codestral.mistral.ai/v1/chat/completions" #Corrected URL
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json",
        "Accept": "application/json"
    }
    response = requests.post(url, headers=headers, data=json.dumps(data))
    return response.json() if response.status_code == 200 else f"Error: {response.status_code}, {response.text}"

def call_fim_endpoint(data, api_key=api_key):
    url = "https://codestral.mistral.ai/v1/fim/completions" #Corrected URL
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json",
        "Accept": "application/json"
    }
    response = requests.post(url, headers=headers, data=json.dumps(data))
    return response.json() if response.status_code == 200 else f"Error: {response.status_code}, {response.text}"

Understanding the Endpoints

Fill-in-the-Middle (FIM) Endpoint:

Generates code to fill the gap between a prompt and an optional suffix.

• URL: https://codestral.mistral.ai/v1/fim/completions
• Parameters: prompt, suffix (optional), stop (optional)

Example:

prompt = "def fibonacci(n: int):"
suffix = "n = int(input('Enter a number: '))\nprint(fibonacci(n))"
data = {"model": "codestral-latest", "prompt": prompt, "suffix": suffix, "temperature": 0}
response = call_fim_endpoint(data)

Instruct Endpoint:

Uses instructions to guide code generation.

• URL: https://codestral.mistral.ai/v1/chat/completions
• Parameters: prompt, temperature (optional), max_tokens (optional)

Example:

import requests
import json

api_key = 'INSERT YOUR API KEY HERE'

def call_chat_endpoint(data, api_key=api_key):
    url = "https://codestral.mistral.ai/v1/chat/completions" #Corrected URL
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json",
        "Accept": "application/json"
    }
    response = requests.post(url, headers=headers, data=json.dumps(data))
    return response.json() if response.status_code == 200 else f"Error: {response.status_code}, {response.text}"

def call_fim_endpoint(data, api_key=api_key):
    url = "https://codestral.mistral.ai/v1/fim/completions" #Corrected URL
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json",
        "Accept": "application/json"
    }
    response = requests.post(url, headers=headers, data=json.dumps(data))
    return response.json() if response.status_code == 200 else f"Error: {response.status_code}, {response.text}"

Advanced Usage

• Rate Limits: codestral.mistral.ai has limits of 30 requests per minute and 2000 per day; api.mistral.ai has 200 requests per second per workspace. Implement retry logic using Python's time library to handle rate limits.
• Error Handling: Handle common errors (401, 429, 500) using appropriate error codes. Retry logic is beneficial for transient errors.
• Customizing Output: Adjust parameters like prompt and temperature to fine-tune the generated code.

Integration

Codestral integrates with IDEs (VS Code, JetBrains) via plugins like Continue.dev. You can also create custom scripts. Here's an example for generating test functions:

prompt = "def fibonacci(n: int):"
suffix = "n = int(input('Enter a number: '))\nprint(fibonacci(n))"
data = {"model": "codestral-latest", "prompt": prompt, "suffix": suffix, "temperature": 0}
response = call_fim_endpoint(data)

Best Practices

• Clear Prompts: Use precise and unambiguous prompts for optimal results.
• Iterative Refinement: Experiment and refine your prompts for better code generation.
• Responsible Use: Use the API ethically and legally, avoiding malicious code generation.

Conclusion

This guide provides a practical introduction to the Codestral API. Experiment and integrate it into your workflow to enhance your development process. For more on Mistral, explore the Mistral 7B tutorial and the guide to working with the Mistral large model.

The above is the detailed content of Codestral API Tutorial: Getting Started With Mistral's API. For more information, please follow other related articles on the PHP Chinese website!