basicrag 0.1.0

Lightweight RAG pipeline for document question answering

RAGlite - Lightweight RAG Pipeline

RAGlite is a simple yet powerful Python package for building Retrieval-Augmented Generation (RAG) pipelines. With just a few lines of code, you can load documents, generate embeddings, and query information using state-of-the-art LLMs.

---

🚀 Features

• ✅ Multi-format support: PDF, DOCX, TXT, and URLs
• ✅ Customizable chunking: Control chunk size and overlap
• ✅ Multiple LLM providers: Groq, Gemini, OpenAI
• ✅ Custom prompts: Bring your own prompt templates
• ✅ Progress logging: Visual console feedback at each step

---

📦 Installation

pip install raglite

---

⚡ Quick Start

from raglite import RAGPipeline

# Initialize pipeline
rag = RAGPipeline(
    llm_provider="groq",
    llm_model="llama3-70b-8192",
    api_key="YOUR_API_KEY"
)

# Load and process data
rag.load_data("https://en.wikipedia.org/wiki/Large_language_model")
rag.fit()

# Query the document
response = rag.query("What are large language models?")
print(response)

---

⚙️ Configuration Options

Parameter	Default	Description
llm_provider	"groq"	LLM provider: "groq", "gemini", "openai"
llm_model	"llama3-70b-8192"	Model name
api_key	None	Provider API key
chunk_size	1000	Text chunk size in characters
chunk_overlap	100	Chunk overlap in characters
top_k	3	Number of chunks to retrieve for context
embedding_model	"sentence-transformers/all-MiniLM-L6-v2"	Sentence embedding model
custom_prompt	None	Custom prompt template

---

✏️ Custom Prompt Templates

Customize the prompt using {context} and {query} placeholders:

custom_prompt = """[INSTRUCTIONS]
Answer the question using ONLY the provided context. 
If unsure, respond "I don't know".

[CONTEXT]
{context}

[QUESTION]
{query}

[ANSWER]
"""

rag = RAGPipeline(
    chunk_size=500,
    chunk_overlap=50,
    top_k=5,
    llm_provider="gemini",
    llm_model="gemini-pro",
    api_key="YOUR_GEMINI_KEY",
    custom_prompt=custom_prompt
)

---

🤖 Supported Models

RAGlite is compatible with any model available through supported providers like Groq, Gemini, and OpenAI, as long as the provider's API supports it.

Below are popular examples, but you're free to use any valid model name:

Groq

• llama3-70b-8192
• llama3-8b-8192
• mixtral-8x7b-32768
• gemma2-9b-it

Gemini

• gemini-pro
• gemini-1.5-pro-latest
• gemini-1.0-pro

OpenAI

• gpt-4-turbo
• gpt-3.5-turbo
• gpt-4o

Tip: You can specify any model string supported by your selected provider using the llm_model parameter in the RAGPipeline.

---

🔧 Advanced Usage

📁 Local File Processing

rag.load_data("research_paper.pdf")
rag.load_data("contract.docx")
rag.load_data("notes.txt")

📚 Multiple Documents

rag = RAGPipeline(...)

# Load multiple sources
rag.load_data("https://example.com/article")
rag.load_data("data/document1.pdf")
rag.load_data("data/notes.txt")

rag.fit()

🔄 Changing Embedding Model

rag = RAGPipeline(
    embedding_model="sentence-transformers/all-mpnet-base-v2",
    # other params...
)

---

🖥️ Command Line Interface (CLI)

RAGlite includes a simple CLI for quick RAG queries:

raglite query "What is the main topic?" \
  --file document.pdf \
  --provider groq \
  --model llama3-70b-8192 \
  --api-key YOUR_KEY

---

🛠 Troubleshooting

Common Issues

• Missing dependencies

  pip install -r requirements.txt
  

• API key not set Set it in code or in a .env file:

  GROQ_API_KEY=your_api_key
  GEMINI_API_KEY=your_api_key
  OPENAI_API_KEY=your_api_key
  

• Unsupported file format Currently supported formats: .pdf, .docx, .txt, and URLs

---

🤝 Contributing

We welcome contributions! To set up your development environment:

1. Clone the repo

2. Create a virtual environment

3. Install dependencies:

   pip install -r requirements.txt
   

4. Start coding!

---

📄 License

MIT License. See LICENSE for details.