Skip to content

Latest commit

 

History

History
249 lines (185 loc) · 5.91 KB

configuration.md

File metadata and controls

249 lines (185 loc) · 5.91 KB
Raw

RAGVenture Configuration Guide

This guide explains how to configure RAGVenture for optimal performance.

Environment Variables

RAGVenture is designed to run completely free, with no API keys required! However, you can optionally use environment variables for custom configurations.

Create a .env file in your project root to configure RAGVenture:

# Required for text generation
HUGGINGFACE_TOKEN="your-token-here"  # Get from huggingface.co

# Optional: LangChain tracking (for debugging)
LANGCHAIN_TRACING_V2=true
LANGCHAIN_ENDPOINT="https://api.smith.langchain.com"
LANGCHAIN_API_KEY="your-langsmith-api-key"
LANGCHAIN_PROJECT="your-project-name"

# Optional: Market analysis configuration
MARKET_DATA_SOURCES="gartner,idc,crunchbase"
MARKET_ANALYSIS_REGION="north_america"
MARKET_CONFIDENCE_THRESHOLD="0.8"

# Optional: Model configuration
MODEL_TEMPERATURE="0.7"  # Controls creativity (0.0-1.0)
MODEL_MAX_LENGTH="1024"  # Maximum output length
MODEL_TOP_P="0.9"       # Nucleus sampling parameter

Model Configuration

RAGVenture uses completely free, locally-running models by default:

Text Generation Models

By default, RAGVenture uses GPT-2 which:

  • Runs completely locally
  • Requires no API key
  • Has no usage costs
  • Provides good performance for idea generation

You can configure the model behavior:

from rag_startups.config import ModelConfig

# Default configuration
config = ModelConfig(
    model_name="gpt2",      # Free, locally-running model
    temperature=0.7,        # Higher = more creative, Lower = more focused
    max_length=1024,        # Maximum token length for generation
    top_p=0.9,             # Controls diversity of outputs
    num_return_sequences=1
)

# Example: More creative outputs
creative_config = ModelConfig(
    model_name="gpt2",
    temperature=0.9,
    top_p=0.95
)

# Example: More focused outputs
focused_config = ModelConfig(
    model_name="gpt2",
    temperature=0.5,
    top_p=0.8
)

# Optional: Use a different local model
config = ModelConfig(
    model_name="gpt2-medium",  # Larger model, still free
    temperature=0.7
)

Embedding Models

We use Sentence Transformers for embeddings, which also:

  • Run completely locally
  • Require no API key
  • Have no usage costs
  • Provide excellent embedding quality
from rag_startups.core.embeddings import get_embedding_model

# Default model (all-MiniLM-L6-v2)
embeddings = get_embedding_model()

# Custom model configuration
embeddings = get_embedding_model(
    model_name="all-mpnet-base-v2",  # Alternative model
    device="cuda"  # Use GPU if available
)

CLI Configuration

Configure the CLI behavior through command-line arguments:

# Control idea generation creativity
python -m src.rag_startups.cli generate-all "AI" --temperature 0.8

# Adjust market analysis
python -m src.rag_startups.cli generate-all "fintech" --market

# Show relevant examples
python -m src.rag_startups.cli generate-all "edtech" --print-examples

# Use custom data source
python -m src.rag_startups.cli generate-all "biotech" --file "custom_data.json"

Performance Optimization

Memory Usage

Control memory usage through these settings:

# Optional: Memory optimization
CHUNK_SIZE="1000"           # Number of records to process at once
MAX_WORKERS="4"             # Number of parallel workers
CACHE_SIZE_MB="512"         # Maximum cache size in MB

GPU Acceleration

Enable GPU acceleration if available:

# Optional: GPU configuration
USE_GPU="true"              # Enable GPU acceleration
CUDA_VISIBLE_DEVICES="0"    # Specify GPU device(s)

Caching Configuration

Configure the caching system:

from rag_startups.utils.cache import configure_cache

# Configure cache settings
configure_cache(
    backend="redis",           # 'redis' or 'local'
    ttl_seconds=3600,         # Cache expiry time
    max_size_mb=512,          # Maximum cache size
    compression=True          # Enable compression
)

Logging Configuration

Control logging behavior:

from rag_startups.utils.logging import configure_logging

# Configure logging
configure_logging(
    level="INFO",            # Logging level
    file="rag_startups.log", # Log file path
    format="detailed",       # Log format
    rotation="1 day"        # Log rotation
)

Docker Configuration

When running in Docker, first make sure your .env file contains all necessary variables:

# Required
HUGGINGFACE_TOKEN="your-token-here"

# Optional
MODEL_TEMPERATURE=0.7
USE_GPU=true
MARKET_DATA_SOURCES=gartner,idc
CACHE_SIZE_MB=512

Then reference these variables in your docker-compose.yml:

environment:
  # Required: Reference token from .env file
  - HUGGINGFACE_TOKEN=${HUGGINGFACE_TOKEN}

  # Optional: Reference other settings from .env
  - MODEL_TEMPERATURE=${MODEL_TEMPERATURE:-0.7}  # Default if not set
  - USE_GPU=${USE_GPU:-true}
  - MARKET_DATA_SOURCES=${MARKET_DATA_SOURCES:-gartner,idc}
  - CACHE_SIZE_MB=${CACHE_SIZE_MB:-512}

This way, all sensitive information like API tokens stays in your .env file (which should be in .gitignore), while docker-compose.yml can be safely committed to version control.

Configuration Precedence

Configuration values are loaded in this order (later overrides earlier):

  1. Default values in code
  2. Environment variables
  3. Configuration file settings
  4. Command-line arguments

Configuration File

You can also use a configuration file (config.yml):

model:
  name: gpt2              # Free, locally-running model
  temperature: 0.7
  max_length: 1024
  top_p: 0.9

market_analysis:
  data_sources:
    - gartner
    - idc
    - crunchbase
  region: north_america
  confidence_threshold: 0.8

cache:
  backend: redis
  ttl_seconds: 3600
  max_size_mb: 512

logging:
  level: INFO
  file: rag_startups.log
  format: detailed
  rotation: 1 day

Load the configuration file:

from rag_startups.config import load_config

config = load_config("config.yml")