🎯 BestShot: The best few shots with LLMs

Ever wished your AI model had a better memory? Meet BestShot - the simple yet powerful library for managing and retrieving few-shot examples with style! 🧠✨

🌟 Features

• 🚀 Lightning Fast: Both sync and async implementations for maximum flexibility
• 🎮 Easy to Use: Simple, intuitive API for managing your AI's example database
• 🔄 Structured Output: Support for structured outputs

🚀 Quick Start

from sentence_transformers import SentenceTransformer # Can also use OpenAI, etc.
from best_shot.client import BestShot
from best_shot.embed.transformers import TransformersEmbedder
from best_shot.store.memory import MemoryStore

# Create a BestShot client
client = BestShot(
    embed=TransformersEmbedder(model=SentenceTransformer("all-MiniLM-L6-v2")),
    store=MemoryStore()
)

# Add some examples
client.add(
    inputs="How do I make a pizza?",
    outputs="1. Make the dough 2. Add toppings 3. Bake at 450°F"
)

# Find similar examples
results = client.list("What's the recipe for pizza?", limit=1)
for shot, similarity in results:
    print(f"Found match (similarity: {similarity:.2f}):")
    print(f"Q: {shot.inputs}")
    print(f"A: {shot.outputs}")

🔧 Installation

pip install best-shot
rye add best-shot
poetry add best-shot

🎮 Usage Examples

Working with Structured Output I/O

# Add structured data
client.add(
    inputs={"type": "greeting", "language": "English"},
    outputs={"text": "Hello, world!"}
)

# Search with similar inputs
results = client.list({"type": "greeting", "language": "English"})

Async Support

from best_shot.async_client import AsyncBestShot

client = AsyncBestShot(embed=async_embedder, store=async_store)

# Add examples asynchronously
await client.add(
    inputs="What's the weather like?",
    outputs="I don't have access to real-time weather data."
)

# Search asynchronously
results = await client.list("How's the weather today?", limit=1)

Using LiteLLM for Embeddings

from functools import partial
from litellm import aembedding
from fewshot import AsyncBestShot
from best_shot.embed.litellm import AsyncLiteLLMEmbedder

client = AsyncBestShot(
    embed=AsyncLiteLLMEmbedder(
        embed=partial(aembedding, model="...", **kwargs),
    ),
    store=MemoryStore()
)

🛠️ Core Components

• Shot: The fundamental unit representing an input-output pair with a unique ID (you can use your own ID or let BestShot hash the inputs)
• Embedder: Converts inputs into vector embeddings for similarity search
• Store: Manages storage and retrieval of examples
• Client: Ties everything together with a clean, simple interface

💡 Use Cases

• 🤖 Enhance your chatbot with dynamic example retrieval
• 📚 Build a self-improving knowledge base
• 🎯 Implement context-aware few-shot learning
• 🧪 Test and experiment with different few-shot strategies

🤝 Contributing

We love contributions! Feel free to:

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

📝 License

MIT License - feel free to use it in your projects!

Made with ❤️ by developers who believe in the power of learning from examples.

Remember: The best AI is the one that learns from experience! 🌟