import streamlit as st # Model Configuration #[model] TYPE = "chat" # so that it uses the chat history SELECTED_MODEL = "meta-llama/llama-3-1-70b-instruct" # Pick a model_id, you can find them here https://dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/fm-api-model-ids.html?context=wx&audience=wdp #SELECTED_MODEL = "meta-llama/llama-3-405b-instruct" #SELECTED_MODEL = "mistralai/mistral-large" #SELECTED_MODEL = "mistralai/mixtral-8x7b-instruct-v01" VERIFY = False # Prompt Configuration #[prompt] SYSTEM_PROMPT = st.secrets["sys_prompt"] PROMPT_TEMPLATE = "llama3-instruct (llama-3 & 3.1) - system" #PROMPT_TEMPLATE = "mistral & mixtral v2 tokenizer - system" # For example "llama3-instruct (llama-3 & 3.1) - user" if you don't use a system prompt. BAKE_IN_PROMPT_SYNTAX = True # Generation Parameters DECODING_METHOD = "greedy" # greedy or sample MAX_NEW_TOKENS = 750 MIN_NEW_TOKENS = 1 REPETITION_PENALTY = 1.0 # LENGTH_PENALTY = {'decay_factor': 1.25, 'start_index': 150} STOP_SEQUENCES = ["<|end_of_text|>",""] # This one is set up for llama models, if you use mistral is the preferred stop_sequence # Additional Parameters - Only active if you pick sampling in decoding method TEMPERATURE = 0.75 TOP_P = 1.0 TOP_K = 50 DISPLAY_CHAT_HISTORY = 1 # 0 to display chat history, 1 to not display chat history # model_family_syntax = { # "llama3-instruct (llama-3 & 3.1) - system": """\n<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>\n\n{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n""", # "llama3-instruct (llama-3 & 3.1) - user": """\n<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n""", # "granite-13b-chat & instruct - system": """\n<|system|>\n{system_prompt}\n<|user|>\n{prompt}\n<|assistant|>\n\n""", # "granite-13b-chat & instruct - user": """\n<|user|>\n{prompt}\n<|assistant|>\n\n""", # "llama2-chat - system": """\n[INST] <>\n{system_prompt}\n<>\n\n{prompt} [/INST] """, # "llama2-chat - user": """\n[INST] {prompt} [/INST] """, # "mistral & mixtral v2 tokenizer - system": """\n[INST]System Prompt:[{system_prompt}]\n\n{prompt} [/INST]\n""", # "mistral & mixtral v2 tokenizer - system segmented": """\n[INST]System Prompt:{system_prompt}[/INST][INST]{prompt} [/INST]\n""", # "mistral & mixtral v2 tokenizer - user": """\n[INST]{prompt} [/INST]\n""" # }