> For the complete documentation index, see [llms.txt](https://docs.arcee.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.arcee.ai/arcee-conductor/arcee-small-language-models/model-capabilities/code-generation.md).

# Code Generation

In this example, you will learn how to use the Arcee coder model for a coding problem.

### Prerequisites

* Python 3.12 or higher
* `httpx` library
* `openai` library
* API key for accessing the Arcee.ai models

### Step 1: Setting Up the Environment

1. Create a new Python virtual environment:

```bash
python -m venv env-openai-client
source env-openai-client/bin/activate  # On Unix/macOS
# or
.\env-openai-client\Scripts\activate  # On Windows
```

2. Install the required packages:

```bash
pip install httpx openai
```

3. Create a file named `api_key.py` containing your API key:

```python
api_key = "your_api_key_here"
```

### Step 2: Initialize the Coder Client

Create a new Jupyter Notebook or Python script and set up the OpenAI client specifically for the Coder model:

```python
import httpx
import os
from openai import OpenAI
from api_key import api_key

endpoint = "https://models.arcee.ai/v1"
model = "coder"  # Arcee's specialized SLM for coding tasks

client = OpenAI(
    base_url=endpoint,
    api_key=api_key,
    http_client=httpx.Client(http2=True)
)
```

### Step 3: Set Up the Response Handler

Create a helper function to handle streaming responses:

```python
def print_streaming_response(response):
    num_tokens = 0
    for message in response:
        if len(message.choices) > 0:
            num_tokens += 1
            print(message.choices[0].delta.content, end="")
    print(f"\n\nNumber of tokens: {num_tokens}")
```

### Step 4: Testing Technical Explanation Capabilities

Test the model's ability to explain complex technical concepts with code examples:

```python
response = client.chat.completions.create(
    model=model,
    messages=[
        {'role': 'user', 
         'content': """Explain the difference between logit-based distillation 
         and hidden state distillation. Show an example for both with Pytorch code, 
         with BERT-Large as the teacher model, and BERT-Base as the student model."""
        }   
    ],
    temperature=0.9,
    stream=True,
    max_tokens=16384
)

print_streaming_response(response)
```

### Step 5: Testing Code Review and Improvement Capabilities

You can use the model to review and improve existing code:

```python
code_example = """
def print_streaming_response(response):
    num_tokens=0
    for message in response:
        if len(message.choices) > 0:
            num_tokens+=1
            print(message.choices[0].delta.content, end="")
    print(f"\\n\\nNumber of tokens: {num_tokens}")
"""

response = client.chat.completions.create(
    model=model,
    messages=[
        {'role': 'user', 
         'content': f"Improve the following code: {code_example}. Explain why your changes are an improvement."
        }   
    ],
    temperature=0.9,
    stream=True,
    max_tokens=2048
)

print_streaming_response(response)
```

### Best Practices for Using the Coder Model

1. **Specific Prompts**:
   * Be specific about the programming language
   * Specify the framework or library you're using
   * Mention any version requirements
   * Include context about the problem you're trying to solve
2. **Code Review Requests**:
   * Include the complete code snippet you want to review
   * Specify what aspects you want to improve (performance, readability, security, etc.)
   * Ask for explanations of suggested improvements
3. **Technical Explanations**:
   * Request specific examples alongside theoretical explanations
   * Ask for comparisons between different approaches
   * Request code snippets that demonstrate the concepts