[tokens]
; Enter one/all of your API key here.
; E.g., OPENAI_API_KEY = sk-xxxxxxx
OPENAI_API_KEY = sk-proj-2JwvyIn7WoKlkbjPOYVWT3BlbkFJnGAk65YAzvPH6cEVQXmr
ANTHROPIC_API_KEY = xxxxx
TOGETHER_API_KEY = xxxxx
; if you use Meta-Llama models, you may need Huggingface token to access.
HUGGINGFACE_TOKEN = xxxxx
VERSION = 1.0.1


[directory]
; Directory for source files.
DOCS_DIR = ./data
; Directory to store embeddings and Langchain documents.
DB_DIR = ./database_store
LOCAL_MODEL_DIR = ./models


; The below parameters are optional to modify:
; --------------------------------------------
[parameters]
; Model name schema: Model Provider|Model Name|Model File. Model File is only valid for GGUF format, set None for other format.

; For example:
; OpenAI|gpt-3.5-turbo|None
; OpenAI|gpt-4|None
; Anthropic|claude-2.0|None
; Together|togethercomputer/llama-2-70b-chat|None
; HuggingFace|TheBloke/Llama-2-70B-chat-GGUF|llama-2-70b-chat.q4_K_M.gguf
; HuggingFace|meta-llama/Llama-2-70b-chat-hf|None

; The full Together.AI model list can be found in the end of this file; We currently only support quantized gguf and the full huggingface local LLMs.
MODEL_NAME = OpenAI|gpt-4-1106-preview|None
; LLM temperature
TEMPURATURE = 0
; Maximum tokens for storing chat history.
MAX_CHAT_HISTORY = 800
; Maximum tokens for LLM context for retrieved information.
MAX_LLM_CONTEXT = 1200
; Maximum tokens for LLM generation.
MAX_LLM_GENERATION = 1000
; Supported embeddings: openAIEmbeddings and hkunlpInstructorLarge.
EMBEDDING_NAME = openAIEmbeddings

; This is dependent on your GPU type.
N_GPU_LAYERS = 100
; this is depend on your GPU and CPU ram when using open source LLMs.
N_BATCH = 512


; The base (small) chunk size for first stage document retrieval.
BASE_CHUNK_SIZE = 100
; Set to 0 for no overlap.
CHUNK_OVERLAP = 0
; The final retrieval (medium) chunk size will be BASE_CHUNK_SIZE * CHUNK_SCALE.
CHUNK_SCALE = 3
WINDOW_STEPS = 3
; The # tokens of window chunk will be BASE_CHUNK_SIZE * WINDOW_SCALE.
WINDOW_SCALE = 18

; Ratio of BM25 retriever to Chroma Vectorstore retriever.
RETRIEVER_WEIGHTS = 0.5, 0.5
; Number of retrieved chunks will range from FIRST_RETRIEVAL_K to 2*FIRST_RETRIEVAL_K due to the ensemble retriever.
FIRST_RETRIEVAL_K = 3
; Number of retrieved chunks will range from SECOND_RETRIEVAL_K to 2*SECOND_RETRIEVAL_K due to the ensemble retriever.
SECOND_RETRIEVAL_K = 3
; Number of windows (large chunks) for the third retriever.
NUM_WINDOWS = 2
; (The third retrieval gets the final chunks passed to the LLM QA chain. The 'k' value is dynamic (based on MAX_LLM_CONTEXT), depending on the number of rephrased questions and retrieved documents.)


[logging]
; If you do not want to enable logging, set enabled to False.
enabled = True
level = INFO
filename = IncarnaMind.log
format = %(asctime)s [%(levelname)s] %(name)s: %(message)s


; Together.AI supported models:

; 0 Austism/chronos-hermes-13b
; 1 EleutherAI/pythia-12b-v0
; 2 EleutherAI/pythia-1b-v0
; 3 EleutherAI/pythia-2.8b-v0
; 4 EleutherAI/pythia-6.9b
; 5 Gryphe/MythoMax-L2-13b
; 6 HuggingFaceH4/starchat-alpha
; 7 NousResearch/Nous-Hermes-13b
; 8 NousResearch/Nous-Hermes-Llama2-13b
; 9 NumbersStation/nsql-llama-2-7B
; 10 OpenAssistant/llama2-70b-oasst-sft-v10
; 11 OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5
; 12 OpenAssistant/stablelm-7b-sft-v7-epoch-3
; 13 Phind/Phind-CodeLlama-34B-Python-v1
; 14 Phind/Phind-CodeLlama-34B-v2
; 15 SG161222/Realistic_Vision_V3.0_VAE
; 16 WizardLM/WizardCoder-15B-V1.0
; 17 WizardLM/WizardCoder-Python-34B-V1.0
; 18 WizardLM/WizardLM-70B-V1.0
; 19 bigcode/starcoder
; 20 databricks/dolly-v2-12b
; 21 databricks/dolly-v2-3b
; 22 databricks/dolly-v2-7b
; 23 defog/sqlcoder
; 24 garage-bAInd/Platypus2-70B-instruct
; 25 huggyllama/llama-13b
; 26 huggyllama/llama-30b
; 27 huggyllama/llama-65b
; 28 huggyllama/llama-7b
; 29 lmsys/fastchat-t5-3b-v1.0
; 30 lmsys/vicuna-13b-v1.3
; 31 lmsys/vicuna-13b-v1.5-16k
; 32 lmsys/vicuna-13b-v1.5
; 33 lmsys/vicuna-7b-v1.3
; 34 prompthero/openjourney
; 35 runwayml/stable-diffusion-v1-5
; 36 stabilityai/stable-diffusion-2-1
; 37 stabilityai/stable-diffusion-xl-base-1.0
; 38 togethercomputer/CodeLlama-13b-Instruct
; 39 togethercomputer/CodeLlama-13b-Python
; 40 togethercomputer/CodeLlama-13b
; 41 togethercomputer/CodeLlama-34b-Instruct
; 42 togethercomputer/CodeLlama-34b-Python
; 43 togethercomputer/CodeLlama-34b
; 44 togethercomputer/CodeLlama-7b-Instruct
; 45 togethercomputer/CodeLlama-7b-Python
; 46 togethercomputer/CodeLlama-7b
; 47 togethercomputer/GPT-JT-6B-v1
; 48 togethercomputer/GPT-JT-Moderation-6B
; 49 togethercomputer/GPT-NeoXT-Chat-Base-20B
; 50 togethercomputer/Koala-13B
; 51 togethercomputer/LLaMA-2-7B-32K
; 52 togethercomputer/Llama-2-7B-32K-Instruct
; 53 togethercomputer/Pythia-Chat-Base-7B-v0.16
; 54 togethercomputer/Qwen-7B-Chat
; 55 togethercomputer/Qwen-7B
; 56 togethercomputer/RedPajama-INCITE-7B-Base
; 57 togethercomputer/RedPajama-INCITE-7B-Chat
; 58 togethercomputer/RedPajama-INCITE-7B-Instruct
; 59 togethercomputer/RedPajama-INCITE-Base-3B-v1
; 60 togethercomputer/RedPajama-INCITE-Chat-3B-v1
; 61 togethercomputer/RedPajama-INCITE-Instruct-3B-v1
; 62 togethercomputer/alpaca-7b
; 63 togethercomputer/codegen2-16B
; 64 togethercomputer/codegen2-7B
; 65 togethercomputer/falcon-40b-instruct
; 66 togethercomputer/falcon-40b
; 67 togethercomputer/falcon-7b-instruct
; 68 togethercomputer/falcon-7b
; 69 togethercomputer/guanaco-13b
; 70 togethercomputer/guanaco-33b
; 71 togethercomputer/guanaco-65b
; 72 togethercomputer/guanaco-7b
; 73 togethercomputer/llama-2-13b-chat
; 74 togethercomputer/llama-2-13b
; 75 togethercomputer/llama-2-70b-chat
; 76 togethercomputer/llama-2-70b
; 77 togethercomputer/llama-2-7b-chat
; 78 togethercomputer/llama-2-7b
; 79 togethercomputer/mpt-30b-chat
; 80 togethercomputer/mpt-30b-instruct
; 81 togethercomputer/mpt-30b
; 82 togethercomputer/mpt-7b-chat
; 83 togethercomputer/mpt-7b
; 84 togethercomputer/replit-code-v1-3b
; 85 upstage/SOLAR-0-70b-16bit
; 86 wavymulder/Analog-Diffusion