aniket47/document-intelligence-chatbot

1# Local Hugging Face Model Settings2EMBEDDING_MODEL = "sentence-transformers/all-MiniLM-L6-v2"  # Fast embedding model3CHAT_MODEL = "google/flan-t5-base"  # Better for summarization and QA tasks4 5# Alternative chat models you can use (just change CHAT_MODEL):6# "google/flan-t5-small" (faster, smaller - 250MB)7# "google/flan-t5-base" (good balance - 990MB) - RECOMMENDED8# "google/flan-t5-large" (better quality, slower - 3GB)9# "facebook/bart-large-cnn" (excellent for summarization but larger)10# "t5-small" (good for summarization, 240MB)11 12# Model Settings13MODEL_MAX_LENGTH = 1000  # Maximum tokens for generation14TEMPERATURE = 0.7  # Creativity (0.0 = deterministic, 1.0 = very creative)15USE_CUDA = True  # Set to False if you don't have GPU16DEVICE = "cpu"  # "auto", "cuda", "cpu"17MODEL_CACHE_DIR = "./models"  # Local directory to cache downloaded models18 19# Document Processing Settings20CHUNK_SIZE = 100021CHUNK_OVERLAP = 20022 23# Vector Store Settings24SIMILARITY_THRESHOLD = 0.125MAX_SEARCH_RESULTS = 526 27# Web Search Settings28WEB_SEARCH_RESULTS = 529WEB_SEARCH_TIMEOUT = 1030 31# Query Routing Settings32WEB_SEARCH_CONFIDENCE_THRESHOLD = 0.633DOCUMENT_SEARCH_CONFIDENCE_THRESHOLD = 0.734HYBRID_THRESHOLD = 0.335 36# Fallback Settings (if local OpenAI models are not available)37USE_SENTENCE_TRANSFORMERS_FALLBACK = True38FALLBACK_EMBEDDING_MODEL = "all-MiniLM-L6-v2"  # Sentence Transformers model39 40# UI Settings41PAGE_TITLE = "Universal Document Intelligence Chatbot"42LAYOUT = "wide"43 44# File Settings45SUPPORTED_FILE_TYPES = ['pdf']46MAX_FILE_SIZE_MB = 5047 48# Response Settings49MAX_RESPONSE_LENGTH = 200050MAX_SOURCES_DISPLAYED = 3