🔧 LLM Suite

A Python package for streamlined interactions with various Large Language Model (LLM) providers, featuring structured output parsing and template-based prompt management.

Features

• 🤖 Multi-Provider Support: Compatible with OpenAI, Anthropic, Ollama, Groq, Perplexity, and LMStudio
• 📋 Structured Output: Parse LLM responses directly into Pydantic models
• 📝 Template Management: Organize and reuse prompts with Jinja2 templating
• ⚙️ Configurable: Easily adjust model parameters and settings

🔧 LLM Suite

A Python package for streamlined interactions with various Large Language Model (LLM) providers, featuring structured output parsing and template-based prompt management.

Features

• 🤖 Multi-Provider Support: Compatible with OpenAI, Anthropic, Ollama, Groq, Perplexity, Together, and LMStudio
• 📋 Structured Output: Parse LLM responses directly into Pydantic models using Instructor
• 📝 Template Management: Organize and reuse prompts with Jinja2 templating
• 🖼️ Image Support: Send images to vision-enabled models
• ⚙️ Configurable: Easily adjust model parameters and settings
• 🔄 CLI Tool: Built-in command line interface for quick interactions

Getting Started

Install the package using pip:

pip install llm-suite

Alternatively, install from the source:

git clone https://github.com/mmysior/llm-suite.git
cd llm-suite
pip install -e .

Configuration

Configuration is handled through environment variables. Copy the .env.example file to .env and update with your API keys:

# LLM settings
DEFAULT_PROVIDER="openai"
DEFAULT_MODEL="gpt-4.1-mini"

DEFAULT_TEMPERATURE=0.7
DEFAULT_TOP_P=1.0
DEFAULT_MAX_TOKENS=2048

# Provider-specific settings
OPENAI_API_KEY=your_openai_key_here
ANTHROPIC_API_KEY=your_anthropic_key_here
PERPLEXITY_API_KEY=your_perplexity_key_here
GROQ_API_KEY=your_groq_key_here
TOGETHER_API_KEY=your_together_key_here
TOGETHER_API_KEY=

OLLAMA_BASE_URL=
LMSTUDIO_BASE_URL=

Usage

Basic Usage

import llmsuite

# Initialize a chat model
llm = llmsuite.init_chat_model()  # Uses DEFAULT_PROVIDER and DEFAULT_MODEL from .env
# Or specify explicitly
llm = llmsuite.init_chat_model(provider="anthropic", model="claude-3-opus-20240229")

# Simple conversation
messages = llm.build_messages(
    text="Explain quantum computing briefly", 
    system_prompt="You are a helpful assistant."
)
response = llm.chat(messages)
print(response)

Structured Output with Pydantic

from pydantic import BaseModel

class MovieRecommendation(BaseModel):
    title: str
    year: int
    why: str

messages = llm.build_messages(
    text="Recommend a sci-fi movie", 
    system_prompt="You recommend movies."
)
result = llm.extract(
    messages=messages,
    schema=MovieRecommendation
)
print(f"Title: {result.title}, Year: {result.year}")

Working with Images

from pathlib import Path

# Send an image to a vision model
messages = llm.build_messages(
    text="What's in this image?",
    image_path=Path("path/to/image.jpg")
)
response = llm.chat(messages)

Prompt Templates

from llmsuite import get_prompt

# Get a prompt template from your templates directory
prompt = get_prompt("my_template")

# Render the template with variables
rendered_prompt = prompt.compile(variable1="value1", variable2="value2")

# Use the rendered prompt
messages = llm.build_messages(text="Query", system_prompt=rendered_prompt)
response = llm.chat(messages)

Creating Prompt Templates

Create Jinja2 templates with YAML frontmatter metadata:

---
type: system
version: 1
author: Your Name
labels: 
    - classification
    - sentiment
tags:
    - example
config:
    temperature: 0.1
    model: gpt-4
---
You are a sentiment analyzer that classifies text as positive, negative, or neutral.

Please analyze the following text:
{{ text }}

Command Line Interface

LLM Suite includes a CLI tool:

# Simple chat
llmsuite chat "What is the capital of France?"

# Using specific model and provider
llmsuite chat "Explain relativity" --model gpt-4 --provider openai --temperature 0.2

# View current configuration
llmsuite config

Advanced Configuration

For more control, you can customize model parameters:

response = llm.chat(
    messages=messages,
    temperature=0.2,
    max_tokens=500
)

License

MIT