Code Interpreter

The code interpreter lets your LLM run Python code directly — perfect for data analysis, file generation, and visualizations. Instead of the LLM just telling you what the code should be, it can actually run it and return the result.

---

What We’ll Build

We’ll make:

1. A data analyst assistant that can run Python.

2. A prompt asking it to create a histogram from a list of numbers.

3. Save the generated chart as a .png file.

---

Step-by-Step

1. Create the LLM Invoker with Code Interpreter Enabled

from gllm_inference.lm_invoker import OpenAILMInvoker

lm_invoker = OpenAILMInvoker(
    model_name="gpt-4.1",
    code_interpreter=True
)

💡 The code_interpreter=True flag gives GPT access to a Python execution sandbox — think of it like a built-in Jupyter Notebook that runs on demand.

---

2. Build the Prompt

We’ll send messages in chat format:

prompt = [
    ("system", ["You are a data analyst. Use the python tool to generate a file."]),
    ("user", ["Show a histogram of the following data: [1, 2, 1, 4, 1, 2, 4, 2, 3, 1]"]),
]

• System message → tells GPT its role (a data analyst).

• User message → our request (make a histogram).

---

3. Invoke the LLM

import asyncio

response = asyncio.run(lm_invoker.invoke(prompt))

Here:

• The LLM might write Python code to plot the histogram.

• The code_interpreter runs that code.

• Any files created (like .png) will be attached to response.code_exec_results.

---

4. Save the Generated File

If the LLM produced a file, we can save it locally:

if response.code_exec_results:
    response.code_exec_results[0].output[0].write_to_file("histogram.png")

💡 write_to_file() takes the output from the interpreter and saves it to disk.

---

Example Run

Prompt:

Show a histogram of the following data: [1, 2, 1, 4, 1, 2, 4, 2, 3, 1]

Behind the scenes:

1. GPT writes Python code using matplotlib.

2. The code interpreter runs it.

3. A histogram image is generated.

4. The file is saved as histogram.png.

Result:

(histogram.png saved in your project folder)

📊 Opening it will show a bar chart of the values.

---

Summary Table

Component	Purpose
code_interpreter=True	Enables Python execution inside GPT
prompt	Defines the role and request
invoke()	Sends the request to GPT
code_exec_results	Holds the output files from the Python run
write_to_file()	Saves the output locally