example.env

# Generic
MODEL_N_CTX=2048
MODEL_MAX_TOKENS=1024
MODEL_STOP=[STOP]
MODEL_TEMP=0.5
MODEL_REPEAT_PENALTY=1.0 
MODEL_TOP_K=40 
MODEL_TOP_P=0.95
MODEL_NO_REPEAT_NGRAM_SIZE=6
MODEL_N_PREDICT=10000
#MODEL_N_PREDICT=-1

MODEL_VERBOSE=true

MODEL_N_THREADS=16
MODEL_N_BATCH=1000
N_GPU_LAYERS=4
USE_MLOCK=false
# keep model loaded in memory
MODEL_USE_MMAP=false

# LlamaCpp or HF
TEXT_EMBEDDINGS_MODEL_TYPE=HF  
TEXT_EMBEDDINGS_MODEL=sentence-transformers/all-MiniLM-L6-v2

# Ingestion
PERSIST_DIRECTORY=db
COLLECTION_NAME=test
DOCUMENTS_DIRECTORY=source_documents
# all-MiniLM-L6-v2 truncates at 256 length of input
INGEST_CHUNK_SIZE=256    
INGEST_CHUNK_OVERLAP=50
INGEST_N_THREADS=8

# Generation # GPT4All or LlamaCpp
MODEL_TYPE=LlamaCpp 
MODEL_PATH=models/WizardLM-7B-uncensored.ggmlv3.q6_K.bin
#stuff or refine or betterstuff or betterrefine   ##### map_reduce # map-rerank 
CHAIN_TYPE=betterrefine
# option helps prevent the model from generating repetitive or monotonous text. A higher value (e.g., 1.5) will penalize repetitions more strongly, while a lower value (e.g., 0.9) will be more lenient. The default value is 1.1.
# document retrieval
# How many documents to retrieve from the db
N_RETRIEVE_DOCUMENTS=100 
# How many documents to forward to the LLM, chosen among those retrieved
N_FORWARD_DOCUMENTS=5