Optimization

.gitignore

@@ -1,4 +1,5 @@
 .env
 __pycache__
 documents
-vecstore
+vecstore
+gDriveDocs

README.md

@@ -34,8 +34,8 @@ docker run --rm -d -p 7860:7860 --env-file ./.env arslan2k12/ttyd_arslanbot
 
 Contents of `.env` file:
 ```
-TTYD_MODE=personalBot_john
-# replace john with your name - use only small alphabets, no special characters
+TTYD_MODE=personalBot_John
+#replace John with your name - use only alphabets, no special characters
 
 GDRIVE_FOLDER_URL=https://drive.google.com/drive/folders/1ce1n1kleS1FOotdcu5joXeSRu_xnHjDt
 # replace with your Google Drive folder URL that has all your knowledge base files (.pdf, .docs, .txt) - make sure this folder is publicly accessible (everyone with the link)

app.py

@@ -1,4 +1,3 @@
-import gdown
 from dotenv import load_dotenv
 import datetime
 import openai
@@ -38,16 +37,16 @@ from ttyd_consts import *
 ###############################################################################################
 
 load_dotenv()
+TTYD_MODE = os.getenv("TTYD_MODE",'')
+
 
 # select the mode when starting container - modes options are in ttyd_consts.py
-if (os.getenv("TTYD_MODE",'')).split('_')[0]=='personalBot':
+if TTYD_MODE.split('_')[0]=='personalBot':
     mode = mode_arslan
-    gDriveUrl = (os.getenv("GDRIVE_FOLDER_URL",'')).replace('?usp=sharing','')
-    # output folder of googe drive folder will be taken as input dir of personalBot
-    gdown.download_folder(url=gDriveUrl, output=mode.inputDir, quiet=True)
-    if os.getenv("TTYD_MODE",'')!='personalBot_arslan':
-        mode.title=''
-        mode.welcomeMsg=''
+    if TTYD_MODE!='personalBot_arslan':
+        user = TTYD_MODE.split('_')[1]
+        mode.title='## Talk to '+user
+        mode.welcomeMsg= welcomeMsgUser(user)
 
 elif os.getenv("TTYD_MODE",'')=='nustian':
     mode = mode_nustian
@@ -57,7 +56,7 @@ else:
 
 if mode.type!='userInputDocs':
     # local vector store as opposed to gradio state vector store, if we the user is not uploading the docs
-    vsDict_hard = localData_vecStore(getPersonalBotApiKey(), inputDir=mode.inputDir, file_list=mode.file_list, url_list=mode.url_list)
+    vsDict_hard = localData_vecStore(getPersonalBotApiKey(), inputDir=mode.inputDir, file_list=mode.file_list, url_list=mode.url_list, gGrUrl=mode.gDriveFolder)
 
 ###############################################################################################
 
@@ -133,9 +132,9 @@ def uiData_vecStore(userFiles, userUrls, api_key_st, vsDict_st={}, progress=gr.P
     return vsDict_st, src_str, *[x.update(interactive=False) for x in [data_ingest_btn, upload_fb]], urls_tb.update(interactive=False, placeholder='')
 
 # initialize chatbot function sets the QA Chain, and also sets/updates any other components to start chatting. updateQaChain function only updates QA chain and will be called whenever Adv Settings are updated.
-def initializeChatbot(temp, k, modelName, stdlQs, api_key_st, vsDict_st, progress=gr.Progress()):
+def initializeChatbot(temp, k, modelNameDD, stdlQs, api_key_st, vsDict_st, progress=gr.Progress()):
     progress(0.1, waitText_initialize)
-    chainTuple = updateQaChain(temp, k, modelName, stdlQs, api_key_st, vsDict_st)
+    chainTuple = updateQaChain(temp, k, modelNameDD, stdlQs, api_key_st, vsDict_st)
     qa_chain_st = chainTuple[0]
     progress(0.5, waitText_initialize)
     #generate welcome message
@@ -147,7 +146,7 @@ def initializeChatbot(temp, k, modelName, stdlQs, api_key_st, vsDict_st, progres
     print('Chatbot initialized at ', datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S'))
 
     return qa_chain_st, chainTuple[1], btn.update(interactive=True), initChatbot_btn.update('Chatbot ready. Now visit the chatbot Tab.', interactive=False)\
-        , status_tb.update(), gr.Tabs.update(selected='cb'), chatbot.update(value=[('Hi', welMsg)])
+        , status_tb.update(), gr.Tabs.update(selected='cb'), chatbot.update(value=[('', welMsg)])
 
 # just update the QA Chain, no updates to any UI
 def updateQaChain(temp, k, modelNameDD, stdlQs, api_key_st, vsDict_st):
@@ -156,15 +155,15 @@ def updateQaChain(temp, k, modelNameDD, stdlQs, api_key_st, vsDict_st):
     
     if api_key_st['service']=='openai':
         if not 'openai' in modelNameDD:
-            modelNameDD = 'gpt-3.5-turbo (openai)'  # default model for openai
+            modelNameDD = changeModel(modelNameDD, OaiDefaultModel)
         llm = getOaiLlm(temp, modelNameDD, api_key_st)
     elif api_key_st['service']=='watsonx':
         if not 'watsonx' in modelNameDD:
-            modelNameDD = 'meta-llama/llama-2-70b-chat (watsonx)' # default model for watsonx
+            modelNameDD = changeModel(modelNameDD, WxDefaultModel)
         llm = getWxLlm(temp, modelNameDD, api_key_st)
     elif api_key_st['service']=='bam':
         if not 'bam' in modelNameDD:
-            modelNameDD = 'meta-llama/llama-2-70b-chat (bam)' # default model for bam
+            modelNameDD = changeModel(modelNameDD, BamDefaultModel)
         llm = getBamLlm(temp, modelNameDD, api_key_st)
     else:
         raise Exception('Error: Invalid or None Credentials')
@@ -261,8 +260,7 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue='orange', secondary_hue='gray
                         temp_sld = gr.Slider(minimum=0, maximum=1, step=0.1, value=0.7, label="Temperature", info='Sampling temperature to use when calling LLM. Defaults to 0.7')
                         k_sld = gr.Slider(minimum=1, maximum=10, step=1, value=mode.k, label="K", info='Number of relavant documents to return from Vector Store. Defaults to 4')
                         model_dd = gr.Dropdown(label='Model Name'\
-                                , choices=model_dd_choices\
-                                , value=model_dd_choices[0], allow_custom_value=True\
+                                , choices=model_dd_choices, allow_custom_value=True\
                                 , info=model_dd_info)
                     stdlQs_rb = gr.Radio(label='Standalone Question', info=stdlQs_rb_info\
                             , type='index', value=stdlQs_rb_choices[1]\

ttyd_consts.py

@@ -1,4 +1,7 @@
 from ibm_watson_machine_learning.foundation_models.utils.enums import ModelTypes
+import os
+from dotenv import load_dotenv
+load_dotenv()
 
 exp_query = 'Generate top 5 questions that I can ask about this data. Questions should be very precise and short, ideally less than 10 words.'
 
@@ -54,10 +57,16 @@ bam_models = sorted(['bigscience/bloom',
  'bigcode/starcoder',
  'google/ul2'])
 
-model_dd_info = 'You can also input any OpenAI model name or BAM model ID.'
+model_dd_info = 'Make sure your credentials are submitted before changing the model. You can also input any OpenAI model name or Watsonx/BAM model ID.' 
 
 model_dd_choices = ['gpt-3.5-turbo (openai)', 'gpt-3.5-turbo-16k (openai)', 'gpt-4 (openai)', 'text-davinci-003 (Legacy - openai)', 'text-curie-001 (Legacy - openai)', 'babbage-002 (openai)'] + [model.value+' (watsonx)' for model in ModelTypes] + [model + ' (bam)' for model in bam_models]
 
+
+OaiDefaultModel = 'gpt-3.5-turbo (openai)'
+WxDefaultModel = 'meta-llama/llama-2-70b-chat (watsonx)'
+BamDefaultModel =  'meta-llama/llama-2-70b-chat (bam)'
+
+
 url_tb_info = 'Upto 100 domain webpages will be crawled for each URL. You can also enter online PDF files.'
 
 url_tb_ph = 'https://example.com, https://another.com, https://anyremotedocument.pdf'
@@ -102,14 +111,22 @@ welcomeMsgArslan = """Summary: The document provides a comprehensive overview of
 
 welcomeMsgDefault = """Hello and welcome! I'm your personal data assistant. Ask me anything about your data and I'll try my best to answer."""
 
+
+def welcomeMsgUser(user):
+    return f"""Hi, Welcome to personal chatbot of {user}. I am trained on the documents {user} has provided me. Ask me anything about {user} and I'll try my best to answer."""
+
+
+gDrFolder=(os.getenv("GDRIVE_FOLDER_URL",'')).replace('?usp=sharing','')
+
 class TtydMode():
-    def __init__(self, name='', title='', type='', dir=None, files=[], urls=[], vis=False, welMsg='', def_k=4):
+    def __init__(self, name='', title='', type='', dir=None, files=[], urls=[], vis=False, welMsg='', def_k=4, gDrFolder=''):
         self.name = name
         self.title = title # markdown title for the top display
         self.type = type # userInputDocs, fixedDocs, personalBot
         self.inputDir=dir
         self.file_list=files
         self.url_list=urls
+        self.gDriveFolder=gDrFolder
         self.uiAddDataVis = vis # load data from user - this will be true for type = userInputDocs
         self.welcomeMsg = welMsg #welcome msg constant - if not provided LLM will generate it
         self.k = def_k # default k docs to retrieve
@@ -118,4 +135,4 @@ class TtydMode():
 
 mode_general = TtydMode(name='general', title=md_title_general, type='userInputDocs', vis=True)
 mode_nustian = TtydMode(name='nustian', title=md_title_nustian, type='fixedDocs', urls=['https://nustian.ca'])
-mode_arslan = TtydMode(name='arslan', title=md_title_arslan, type='personalBot', dir='./documents/', welMsg=welcomeMsgArslan, def_k=8)
+mode_arslan = TtydMode(name='arslan', title=md_title_arslan, type='personalBot', welMsg=welcomeMsgArslan, def_k=8, gDrFolder=gDrFolder)

ttyd_functions.py

@@ -1,5 +1,7 @@
 
 import datetime
+import gradio as gr
+import time
 import uuid
 import openai
 from langchain.embeddings import OpenAIEmbeddings
@@ -19,6 +21,7 @@ from urllib.parse import urlparse
 import mimetypes
 from pathlib import Path
 import tiktoken
+import gdown
 
 from langchain.chat_models import ChatOpenAI
 from langchain import OpenAI
@@ -67,6 +70,8 @@ def getPersonalBotApiKey():
         return getOaiCreds(os.getenv("OPENAI_API_KEY"))
     elif os.getenv("WX_API_KEY") and os.getenv("WX_PROJECT_ID"):
         return getWxCreds(os.getenv("WX_API_KEY"), os.getenv("WX_PROJECT_ID"))
+    elif os.getenv("BAM_API_KEY"):
+        return getBamCreds(os.getenv("BAM_API_KEY"))
     else:
         return {}
     
@@ -240,24 +245,30 @@ def ingestFiles(documents, files_list, prog=None):
             pass
         
         if doc is not None and doc[0].page_content:
-            if prog is not None: prog(1, desc='Loaded file: '+fPath.rsplit('/')[0])
+            if prog is not None: prog(0.9, desc='Loaded file: '+fPath.rsplit('/')[0])
             print('Loaded file:', fPath)
             documents.extend(doc)
     return documents
 
 
-def data_ingestion(inputDir=None, file_list=[], url_list=[], prog=None):
+def data_ingestion(inputDir=None, file_list=[], url_list=[], gDriveFolder='', prog=None):
     documents = []
+    # Ingestion from Google Drive Folder
+    if gDriveFolder:
+        opFolder = './gDriveDocs/'
+        gdown.download_folder(url=gDriveFolder, output=opFolder, quiet=True)
+        files = [str(x) for x in Path(opFolder).glob('**/*')]
+        documents = ingestFiles(documents, files, prog)
     # Ingestion from Input Directory
     if inputDir is not None:
         files = [str(x) for x in Path(inputDir).glob('**/*')]
-        documents = ingestFiles(documents, files)
+        documents = ingestFiles(documents, files, prog)
     if file_list:
         documents = ingestFiles(documents, file_list, prog)
     # Ingestion from URLs - also try https://python.langchain.com/docs/integrations/document_loaders/recursive_url_loader
     if url_list:
         for url in url_list:
-            documents = ingestURL(documents, url, prog=prog)        
+            documents = ingestURL(documents, url, prog=prog)
 
     # Cleanup documents
     for x in documents:
@@ -331,8 +342,8 @@ def getVsDict(embeddingFunc, docs, vsDict={}):
     return vsDict
 
 # used for Hardcoded documents only - not uploaded by user (userData_vecStore is separate function)
-def localData_vecStore(embKey={}, inputDir=None, file_list=[], url_list=[], vsDict={}):
-    documents = data_ingestion(inputDir, file_list, url_list)
+def localData_vecStore(embKey={}, inputDir=None, file_list=[], url_list=[], vsDict={}, gGrUrl=''):
+    documents = data_ingestion(inputDir, file_list, url_list, gGrUrl)
     if not documents:
        raise Exception('Error: No Documents Found')
     docs = split_docs(documents)
@@ -353,3 +364,9 @@ def num_tokens_from_string(string, encoding_name = "cl100k_base"):
     num_tokens = len(encoding.encode(string))
     return num_tokens
 
+def changeModel(oldModel, newModel):
+    if oldModel:
+        warning = 'Credentials not found for '+oldModel+'. Using default model '+newModel
+        gr.Warning(warning)
+        time.sleep(1)
+    return newModel