config.yaml

DEVICE: cpu

CHUNK_SIZE: 1000
CHUNK_OVERLAP: 50
SEPARATORS: ["\n\n", "\n", ". ", " ", ""]

TEXT_SPLIT_MODE: simple

# TEXT_SPLIT_MODE: parent_document
# PARENT_DOCS_PATH: ./data/parent_docs.pkl

MODELS_DIR: ./models

EMBEDDINGS_PATH: all-mpnet-base-v2
# EMBEDDINGS_PATH: bge-small-en-v1.5

# VECTORDB_TYPE: faiss
# VECTORDB_PATH: vectordb/faiss
VECTORDB_TYPE: chroma
VECTORDB_PATH: vectordb/chroma

# USE_CTRANSFORMERS: True
# LLM_PATH: llama-2-7b-chat.Q4_K_M.gguf
# PROMPT_TYPE: llama

# USE_CTRANSFORMERS: True
# LLM_PATH: mistral-7b-instruct-v0.2.Q4_K_M.gguf
# PROMPT_TYPE: mistral

# USE_CTRANSFORMERS: True
# LLM_PATH: zephyr-7b-beta.Q4_K_M.gguf
# PROMPT_TYPE: zephyr

USE_CTRANSFORMERS: False
LLM_PATH: gemma-2b-it-q4_k_m.gguf
PROMPT_TYPE: gemma

LLM_CONFIG:
  MAX_NEW_TOKENS: 512
  TEMPERATURE: 0.2
  REPETITION_PENALTY: 1.1
  CONTEXT_LENGTH: 2048

BASE_RETRIEVER_CONFIG:
  SEARCH_K: 4

# RERANKER_PATH: tart-full-flan-t5-xl
# RERANKER_TYPE: tart
RERANKER_PATH: bge-reranker-base
RERANKER_TYPE: bge

RERANK_RETRIEVER_CONFIG:
  SEARCH_K: 10
  TOP_N: 4

COMPRESSION_RETRIEVER_CONFIG:
  SEARCH_K: 10
  SIMILARITY_THRESHOLD: 0.5

PROPOSITIONIZER_PATH: propositionizer-wiki-flan-t5-large
PROPOSITIONIZER_CONFIG:
  CHUNK_SIZE: 1000
  CHUNK_OVERLAP: 0