LLM Generation

This tutorial demonstrates how to use Hypster with the llm package for managing different LLM configurations. We'll create a simple example showing how to switch between models and adjust generation parameters.

Prerequisites:

pip install llm

Configurable LLM

import os
import llm
from hypster import HP, instantiate

def llm_config(hp: HP):
    model_name = hp.select(["gpt-4o-mini", "gpt-4o"], name="model_name")
    temperature = hp.float(0.0, name="temperature", min=0.0, max=1.0)
    max_tokens = hp.int(256, name="max_tokens", max=2048)

    return {
        "model_name": model_name,
        "temperature": temperature,
        "max_tokens": max_tokens
    }

def generate(prompt: str,
             model_name: str,
             temperature: float,
             max_tokens: int) -> str:
    model = llm.get_model(model_name)
    return model.prompt(prompt, temperature=temperature, max_tokens=max_tokens)


os.environ["OPENAI_API_KEY"] = "..."

# Create configurations for different use cases
default_config = instantiate(llm_config)
creative_config = instantiate(llm_config, values={
    "model_name": "gpt-4o",
    "temperature": 1.0
})

# Example prompts
prompt1 = "Explain what machine learning is in 5 words."
prompt2 = "Write a haiku about AI in 17 syllables."

# Generate responses with different configurations
print("Default Configuration (Balanced):")
print(generate(prompt1, **default_config))

print("Creative Configuration (Higher Temperature):")
print(generate(prompt2, **creative_config))

This example demonstrates:

1. Simple model configuration with Hypster

2. Easy model switching using llm

3. Adjustable generation parameters (temperature, max_tokens)

4. Different configurations for different use cases

5. Explicit parameter naming with name="..." arguments

6. Using instantiate() for configuration execution