Create genparam_old_backup.py
Browse files- genparam_old_backup.py +55 -0
genparam_old_backup.py
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from secretsload import load_stsecrets
|
3 |
+
|
4 |
+
credentials = load_stsecrets()
|
5 |
+
|
6 |
+
# Model Configuration
|
7 |
+
ACTIVE_MODEL = 0 # 0 == SELECTED_MODEL_1 and PROMPT_TEMPLATE_1
|
8 |
+
ACTIVE_INDEX = 0 # 0 == VECTOR_INDEX_1
|
9 |
+
|
10 |
+
TYPE = "chat" # so that it uses the chat history
|
11 |
+
SELECTED_MODEL_1 = "meta-llama/llama-3-1-70b-instruct"
|
12 |
+
SELECTED_MODEL_2 = "mistralai/mistral-large"
|
13 |
+
# 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 or switch between the two options up there by adjusting the ACTIVE_MODEL variable between 0 (1) and 1 (2).
|
14 |
+
VERIFY = False
|
15 |
+
|
16 |
+
# Prompt Configuration
|
17 |
+
PROMPT_TEMPLATE_1 = "llama3-instruct (llama-3, 3.1 & 3.2) - system"
|
18 |
+
PROMPT_TEMPLATE_2 = "mistral & mixtral v2 tokenizer - system segmented"
|
19 |
+
# <pick prompt template from model_family_syntax below> For example "llama3-instruct (llama-3 & 3.1) - user" if you don't use a system prompt.
|
20 |
+
BAKE_IN_PROMPT_SYNTAX = True
|
21 |
+
|
22 |
+
# Bot Names
|
23 |
+
BOT_1_NAME = "PATH-er B."
|
24 |
+
BOT_2_NAME = "MOD-ther S."
|
25 |
+
BOT_3_NAME = "SYS-ter V."
|
26 |
+
|
27 |
+
# Avatars
|
28 |
+
BOT_1_AVATAR = "🤖" # Robot for PATH-er B.
|
29 |
+
BOT_2_AVATAR = "🦾" # Mechanical arm for MOD-ther S.
|
30 |
+
BOT_3_AVATAR = "🎭" # Theatre masks for SYS-ter V.
|
31 |
+
USER_AVATAR = "👤" # Keep the existing user avatar
|
32 |
+
|
33 |
+
# Bot Prompts
|
34 |
+
BOT_1_PROMPT = str(st.secrets["system_prompt_1"])
|
35 |
+
BOT_2_PROMPT = str(st.secrets["system_prompt_2"])
|
36 |
+
BOT_3_PROMPT = str(st.secrets["system_prompt_3"])
|
37 |
+
|
38 |
+
# Vector Indexes
|
39 |
+
VECTOR_INDEX_1 = str(st.secrets["vector_index_id_1"])
|
40 |
+
VECTOR_INDEX_2 = str(st.secrets["vector_index_id_2"])
|
41 |
+
|
42 |
+
# Generation Parameters
|
43 |
+
DECODING_METHOD = "greedy" # greedy or sample
|
44 |
+
MAX_NEW_TOKENS = 850
|
45 |
+
MIN_NEW_TOKENS = 1
|
46 |
+
REPETITION_PENALTY = 1.0
|
47 |
+
STOP_SEQUENCES = ["<|end_of_text|>","</s>"] # This one is set up for llama models, if you use mistral </s> is the preferred stop_sequence
|
48 |
+
|
49 |
+
# Additional Parameters - Only active if you pick sampling in decoding method
|
50 |
+
TEMPERATURE = 0.7
|
51 |
+
TOP_P = 1.0
|
52 |
+
TOP_K = 50
|
53 |
+
|
54 |
+
DISPLAY_CHAT_HISTORY = 1 # 0 to not display chat history, 1 to display chat history
|
55 |
+
TOKEN_CAPTURE_ENABLED = 0 # Set to 1 to enable token capture preview in the side_bar, 0 to disable
|