🎯 FewShots: The best few shots with LLMs

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

🌟 Features

• 🎮 Easy to Use: Simple, intuitive API for managing your AI's example database
• 🔄 Structured Output: Support for structured outputs

💡 Use Cases

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

🛠️ Core Components

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

🔧 Installation

pip install few-shots
rye add few-shots
poetry add few-shots

🚀 Quick Start

from few_shots.client import FewShots
from few_shots.embed.openai import OpenAIEmbed
from few_shots.store.memory import MemoryStore # see below for different vectorstores
from few_shots.types import Shot

from openai import OpenAI

shots = FewShots(
    embed=OpenAIEmbed(
        OpenAI().embeddings.create,
        model="...",
        **kwargs,
    ),
    store=MemoryStore()
)

# Works with strings or dictionaries (for structured inputs/outputs)
shots.add(
    inputs=str | dict,
    outputs=str | dict,
    id=str | None # For upserts, FewShots will hash the inputs by default to generate a UUID5, or, bring your own str(ID)
)

# When a user calls your app, you can use the `get` method to retrieve cached, known good examples
shot: Shot | None = shots.get(inputs=...)
if not shot:
    # Get similar examples
    knn_shots = shots.list(inputs=..., limit=10) # default = 5

    for distance, shot in knn_shots:
        print(f"Found match (distance: {distance:.2f}):")
        print(f"Q: {shot.inputs}")
        print(f"A: {shot.outputs}")

    # Use with your LLM
    from few_shots.utils.format import shots_to_messages

    response = openai.chat.completions.create(
        ...,
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            *shots_to_messages(knn_shots),
            {"role": "user", "content": "What's the recipe for pizza?"},
        ]
    )

🎮 Usage Examples

Working with Structured Output I/O

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

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

Using persistent Vector Stores

from few_shots.store.pg import PGStore, AsyncPGStore
from few_shots.store.chroma import ChromaStore, AsyncChromaStore
from few_shots.store.qdrant import QdrantStore, AsyncQdrantStore
from few_shots.store.weaviate import WeaviateStore, AsyncWeaviateStore
from few_shots.store.turbopuffer import TurboPufferStore, AsyncTurboPufferStore # Untested
from few_shots.store.milvus import MilvusStore, AsyncMilvusStore # Untested

# check out the store's .setup method to see how to configure it
# this method creates the table, collection, indexes, etc. and is idempotent

Using OpenAI / LiteLLM for Embeddings

The OpenAIEmbed and AsyncOpenAIEmbed classes are compatible with all OpenAI-compatible SDKs.

from few_shots import AsyncFewShots
from few_shots.embed.openai import OpenAIEmbed, AsyncOpenAIEmbed # Compatible with all OpenAI

from openai import OpenAI

shots = FewShots(
    embed=OpenAIEmbed(
        OpenAI().embeddings.create,
        model="...",
        **kwargs,
    ),
    store=MemoryStore()
)

from litellm import aembedding

shots = AsyncFewShots(
    embed=AsyncOpenAIEmbed(
        aembedding,
        model="...",
        **kwargs,
    ),
    store=MemoryStore()
)

🤝 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! 🌟