Merge branch 'master' of github.com:debanjum/semantic-search into saba/configui

environment.yml

@@ -12,4 +12,5 @@ dependencies:
   - pyyaml=5.*
   - pytest=6.*
   - pillow=8.*
-  - torchvision=0.*
+  - torchvision=0.*
+  - openai=0.*

sample_config.yml

@@ -30,3 +30,9 @@ search-type:
 
   image:
     encoder: "clip-ViT-B-32"
+
+processor:
+  conversation:
+    openai-api-key: null
+    conversation-logfile: "tests/data/.conversation_logs.json"
+    conversation-history: null

src/main.py

@@ -1,5 +1,6 @@
 # Standard Packages
 import sys
+import json
 from typing import Optional
 
 # External Packages
@@ -10,13 +11,15 @@ from fastapi.templating import Jinja2Templates
 
 # Internal Packages
 from src.search_type import asymmetric, symmetric_ledger, image_search
-from src.utils.helpers import get_from_dict
+from src.utils.helpers import get_absolute_path
 from src.utils.cli import cli
-from src.utils.config import SearchType, SearchModels, TextSearchConfig, ImageSearchConfig, SearchConfig
+from src.utils.config import SearchType, SearchModels, TextSearchConfig, ImageSearchConfig, SearchConfig, ProcessorConfig, ConversationProcessorConfig
+from src.processor.conversation.gpt import converse, message_to_prompt
 
 # Application Global State
 model = SearchModels()
 search_config = SearchConfig()
+processor_config = ProcessorConfig()
 app = FastAPI()
 
 # app.mount("/views", StaticFiles(directory="./views"), name="views")
@@ -92,6 +95,20 @@ def regenerate(t: Optional[SearchType] = None):
     return {'status': 'ok', 'message': 'regeneration completed'}
 
 
+@app.get('/chat')
+def chat(q: str):
+    # Load Conversation History
+    conversation_history = processor_config.conversation.conversation_history
+
+    # Converse with OpenAI GPT
+    gpt_response = converse(q, conversation_history, api_key=processor_config.conversation.openai_api_key)
+
+    # Update Conversation History
+    processor_config.conversation.conversation_history = message_to_prompt(q, conversation_history, gpt_response)
+
+    return {'status': 'ok', 'response': gpt_response}
+
+
 def initialize_search(config, regenerate, verbose):
     model = SearchModels()
     search_config = SearchConfig()
@@ -119,6 +136,39 @@ def initialize_search(config, regenerate, verbose):
     return model, search_config
 
 
+def initialize_processor(config, verbose):
+    processor_config = ProcessorConfig()
+
+    # Initialize Conversation Processor
+    processor_config.conversation = ConversationProcessorConfig.create_from_dictionary(config, ('processor', 'conversation'), verbose)
+
+    # Load or Initialize Conversation History from Disk
+    conversation_logfile = processor_config.conversation.conversation_logfile
+    if processor_config.conversation.verbose:
+        print('Saving conversation logs to disk...')
+
+    if conversation_logfile.expanduser().absolute().is_file():
+        with open(get_absolute_path(conversation_logfile), 'r') as f:
+            processor_config.conversation.conversation_history = json.load(f).get('chat', '')
+    else:
+        processor_config.conversation.conversation_history = ''
+
+    return processor_config
+
+
+@app.on_event('shutdown')
+def shutdown_event():
+    if processor_config.conversation.verbose:
+        print('Saving conversation logs to disk...')
+
+    # Save Conversation History to Disk
+    conversation_logfile = get_absolute_path(processor_config.conversation.conversation_logfile)
+    with open(conversation_logfile, "w+", encoding='utf-8') as logfile:
+        json.dump({"chat": processor_config.conversation.conversation_history}, logfile)
+
+    print('Conversation logs saved to disk.')
+
+
 if __name__ == '__main__':
     # Load config from CLI
     args = cli(sys.argv[1:])
@@ -126,6 +176,9 @@ if __name__ == '__main__':
     # Initialize Search from Config
     model, search_config = initialize_search(args.config, args.regenerate, args.verbose)
 
+    # Initialize Processor from Config
+    processor_config = initialize_processor(args.config, args.verbose)
+
     # Start Application Server
     if args.socket:
         uvicorn.run(app, proxy_headers=True, uds=args.socket)

src/processor/conversation/gpt.py

@@ -0,0 +1,70 @@
+# Standard Packages
+import os
+
+# External Packages
+import openai
+
+
+def understand(text, api_key=None, temperature=0.5, max_tokens=100):
+    """
+    Understand user input using OpenAI's GPT
+    """
+    # Initialize Variables
+    openai.api_key = api_key or os.getenv("OPENAI_API_KEY")
+    understand_primer="Extract information from each chat message\n\nremember(memory-type, data);\nmemory-type=[\"companion\", \"notes\", \"ledger\", \"image\", \"music\"]\nsearch(search-type, data);\nsearch-type=[\"google\", \"youtube\"]\ngenerate(activity);\nactivity=[\"paint\",\"write\", \"chat\"]\ntrigger-emotion(emotion);\nemotion=[\"happy\",\"confidence\",\"fear\",\"surprise\",\"sadness\",\"disgust\",\"anger\", \"curiosity\", \"calm\"]\n\nQ: How are you doing?\nA: activity(\"chat\"); trigger-emotion(\"surprise\")\nQ: Do you remember what I told you about my brother Antoine when we were at the beach?\nA: remember(\"notes\", \"Brother Antoine when we were at the beach\"); trigger-emotion(\"curiosity\");\nQ: what did we talk about last time?\nA: remember(\"notes\", \"talk last time\"); trigger-emotion(\"curiosity\");\nQ: Let's make some drawings!\nA: generate(\"paint\"); trigger-emotion(\"happy\");\nQ: Do you know anything about Lebanon?\nA: search(\"google\", \"lebanon\");  trigger-emotion(\"confidence\");\nQ: Find a video about a panda rolling in the grass\nA: search(\"youtube\",\"panda rolling in the grass\");  trigger-emotion(\"happy\"); \nQ: Tell me a scary story\nA: generate(\"write\" \"A story about some adventure\"); trigger-emotion(\"fear\");\nQ: What fiction book was I reading last week about AI starship?\nA: remember(\"notes\", \"read fiction book about AI starship last week\"); trigger-emotion(\"curiosity\");\nQ: How much did I spend at Subway for dinner last time?\nA: remember(\"ledger\", \"last Subway dinner\"); trigger-emotion(\"curiosity\");\nQ: I'm feeling sleepy\nA: activity(\"chat\"); trigger-emotion(\"calm\")\nQ: What was that popular Sri lankan song that Alex showed me recently?\nA: remember(\"music\", \"popular Sri lankan song that Alex showed recently\");  trigger-emotion(\"curiosity\"); \nQ: You're pretty funny!\nA: activity(\"chat\"); trigger-emotion(\"pride\")"
+
+    # Setup Prompt with Understand Primer
+    prompt = message_to_prompt(text, understand_primer, start_sequence="\nA:", restart_sequence="\nQ:")
+
+    # Get Reponse from GPT
+    response = openai.Completion.create(
+        engine="davinci",
+        prompt=prompt,
+        temperature=temperature,
+        max_tokens=max_tokens,
+        top_p=1,
+        frequency_penalty=0.2,
+        presence_penalty=0,
+        stop=["\n"])
+
+    # Extract, Clean Message from GPT's Response
+    story = response['choices'][0]['text']
+    return str(story)
+
+
+def converse(text, conversation_history=None, api_key=None, temperature=0.9, max_tokens=150):
+    """
+    Converse with user using OpenAI's GPT
+    """
+    # Initialize Variables
+    openai.api_key = api_key or os.getenv("OPENAI_API_KEY")
+
+    start_sequence = "\nAI:"
+    restart_sequence = "\nHuman:"
+    conversation_primer = f"The following is a conversation with an AI assistant. The assistant is helpful, creative, clever, and very friendly companion.\n{restart_sequence} Hello, who are you?{start_sequence} Hi, I am an AI conversational companion created by OpenAI. How can I help you today?"
+
+    # Setup Prompt with Primer or Conversation History
+    prompt = message_to_prompt(text, conversation_history or conversation_primer, start_sequence=start_sequence, restart_sequence=restart_sequence)
+
+    # Get Response from GPT
+    response = openai.Completion.create(
+        engine="davinci",
+        prompt=prompt,
+        temperature=temperature,
+        max_tokens=max_tokens,
+        top_p=1,
+        frequency_penalty=0,
+        presence_penalty=0.6,
+        stop=["\n", " Human:", " AI:"])
+
+    # Extract, Clean Message from GPT's Response
+    story = response['choices'][0]['text']
+    return str(story).strip()
+
+
+def message_to_prompt(user_message, conversation_history="", gpt_message=None, start_sequence="\nAI:", restart_sequence="\nHuman:"):
+    """Create prompt for GPT from message"""
+    if gpt_message:
+        return f"{conversation_history}{restart_sequence} {user_message}{start_sequence} {gpt_message}"
+    else:
+        return f"{conversation_history}{restart_sequence} {user_message}{start_sequence}"

src/utils/cli.py

@@ -80,6 +80,15 @@ default_config = {
         'image':
         {
             'encoder': "clip-ViT-B-32"
-        }
+        },
+    },
+    'processor':
+    {
+        'conversation':
+        {
+            'openai-api-key': "",
+            'conversation-logfile': ".conversation_logs.json",
+            'conversation-history': ""
+        },
     }
 }

src/utils/config.py

@@ -93,3 +93,27 @@ class SearchConfig():
     ledger: TextSearchConfig = None
     music: TextSearchConfig = None
     image: ImageSearchConfig = None
+
+
+class ConversationProcessorConfig():
+    def __init__(self, conversation_logfile, conversation_history, openai_api_key, verbose):
+        self.openai_api_key = openai_api_key
+        self.conversation_logfile = conversation_logfile
+        self.conversation_history = conversation_history
+        self.verbose = verbose
+
+    def create_from_dictionary(config, key_tree, verbose):
+        conversation_config = get_from_dict(config, *key_tree)
+        if not conversation_config:
+            return None
+
+        return ConversationProcessorConfig(
+            openai_api_key = conversation_config['openai-api-key'],
+            conversation_history = '',
+            conversation_logfile = Path(conversation_config['conversation-logfile']),
+            verbose = verbose)
+
+
+@dataclass
+class ProcessorConfig():
+    conversation: ConversationProcessorConfig = None

tests/test_chatbot.py

@@ -0,0 +1,63 @@
+# External Packages
+import pytest
+
+# Internal Packages
+from src.processor.conversation.gpt import converse, understand, message_to_prompt
+
+# Input your OpenAI API key to run the tests below
+api_key = None
+
+
+# Test
+# ----------------------------------------------------------------------------------------------------
+def test_message_to_understand_prompt():
+    # Setup
+    understand_primer = "Extract information from each chat message\n\nremember(memory-type, data);\nmemory-type=[\"companion\", \"notes\", \"ledger\", \"image\", \"music\"]\nsearch(search-type, data);\nsearch-type=[\"google\", \"youtube\"]\ngenerate(activity);\nactivity=[\"paint\",\"write\", \"chat\"]\ntrigger-emotion(emotion);\nemotion=[\"happy\",\"confidence\",\"fear\",\"surprise\",\"sadness\",\"disgust\",\"anger\", \"curiosity\", \"calm\"]\n\nQ: How are you doing?\nA: activity(\"chat\"); trigger-emotion(\"surprise\")\nQ: Do you remember what I told you about my brother Antoine when we were at the beach?\nA: remember(\"notes\", \"Brother Antoine when we were at the beach\"); trigger-emotion(\"curiosity\");\nQ: what did we talk about last time?\nA: remember(\"notes\", \"talk last time\"); trigger-emotion(\"curiosity\");\nQ: Let's make some drawings!\nA: generate(\"paint\"); trigger-emotion(\"happy\");\nQ: Do you know anything about Lebanon?\nA: search(\"google\", \"lebanon\");  trigger-emotion(\"confidence\");\nQ: Find a video about a panda rolling in the grass\nA: search(\"youtube\",\"panda rolling in the grass\");  trigger-emotion(\"happy\"); \nQ: Tell me a scary story\nA: generate(\"write\" \"A story about some adventure\"); trigger-emotion(\"fear\");\nQ: What fiction book was I reading last week about AI starship?\nA: remember(\"notes\", \"read fiction book about AI starship last week\"); trigger-emotion(\"curiosity\");\nQ: How much did I spend at Subway for dinner last time?\nA: remember(\"ledger\", \"last Subway dinner\"); trigger-emotion(\"curiosity\");\nQ: I'm feeling sleepy\nA: activity(\"chat\"); trigger-emotion(\"calm\")\nQ: What was that popular Sri lankan song that Alex showed me recently?\nA: remember(\"music\", \"popular Sri lankan song that Alex showed recently\");  trigger-emotion(\"curiosity\"); \nQ: You're pretty funny!\nA: activity(\"chat\"); trigger-emotion(\"pride\")"
+    expected_response = "Extract information from each chat message\n\nremember(memory-type, data);\nmemory-type=[\"companion\", \"notes\", \"ledger\", \"image\", \"music\"]\nsearch(search-type, data);\nsearch-type=[\"google\", \"youtube\"]\ngenerate(activity);\nactivity=[\"paint\",\"write\", \"chat\"]\ntrigger-emotion(emotion);\nemotion=[\"happy\",\"confidence\",\"fear\",\"surprise\",\"sadness\",\"disgust\",\"anger\", \"curiosity\", \"calm\"]\n\nQ: How are you doing?\nA: activity(\"chat\"); trigger-emotion(\"surprise\")\nQ: Do you remember what I told you about my brother Antoine when we were at the beach?\nA: remember(\"notes\", \"Brother Antoine when we were at the beach\"); trigger-emotion(\"curiosity\");\nQ: what did we talk about last time?\nA: remember(\"notes\", \"talk last time\"); trigger-emotion(\"curiosity\");\nQ: Let's make some drawings!\nA: generate(\"paint\"); trigger-emotion(\"happy\");\nQ: Do you know anything about Lebanon?\nA: search(\"google\", \"lebanon\");  trigger-emotion(\"confidence\");\nQ: Find a video about a panda rolling in the grass\nA: search(\"youtube\",\"panda rolling in the grass\");  trigger-emotion(\"happy\"); \nQ: Tell me a scary story\nA: generate(\"write\" \"A story about some adventure\"); trigger-emotion(\"fear\");\nQ: What fiction book was I reading last week about AI starship?\nA: remember(\"notes\", \"read fiction book about AI starship last week\"); trigger-emotion(\"curiosity\");\nQ: How much did I spend at Subway for dinner last time?\nA: remember(\"ledger\", \"last Subway dinner\"); trigger-emotion(\"curiosity\");\nQ: I'm feeling sleepy\nA: activity(\"chat\"); trigger-emotion(\"calm\")\nQ: What was that popular Sri lankan song that Alex showed me recently?\nA: remember(\"music\", \"popular Sri lankan song that Alex showed recently\");  trigger-emotion(\"curiosity\"); \nQ: You're pretty funny!\nA: activity(\"chat\"); trigger-emotion(\"pride\")\nQ: When did I last dine at Burger King?\nA:"
+
+    # Act
+    actual_response = message_to_prompt("When did I last dine at Burger King?", understand_primer, start_sequence="\nA:", restart_sequence="\nQ:")
+
+    # Assert
+    assert actual_response == expected_response
+
+
+# ----------------------------------------------------------------------------------------------------
+@pytest.mark.skipif(api_key is None,
+                    reason="Set api_key variable to your OpenAI API key from https://beta.openai.com/account/api-keys")
+def test_minimal_chat_with_gpt():
+    # Act
+    response = converse("What will happen when the stars go out?", api_key=api_key)
+
+    # Assert
+    assert len(response) > 0
+
+
+# ----------------------------------------------------------------------------------------------------
+@pytest.mark.skipif(api_key is None,
+                    reason="Set api_key variable to your OpenAI API key from https://beta.openai.com/account/api-keys")
+def test_chat_with_history():
+    # Act
+    start_sequence="\nAI:"
+    restart_sequence="\nHuman:"
+
+    conversation_primer = f"The following is a conversation with an AI assistant. The assistant is helpful, creative, clever, and very friendly companion.\n{restart_sequence} Hello, I am testatron. Who are you?{start_sequence} Hi, I am an AI conversational companion created by OpenAI. How can I help you today?"
+    conversation_history = conversation_primer
+
+    response = converse("Can you tell me my name?", conversation_history=conversation_history, api_key=api_key, temperature=0, max_tokens=50)
+
+    # Assert
+    assert len(response) > 0
+    assert "Testatron" in response or "testatron" in response
+
+
+# ----------------------------------------------------------------------------------------------------
+@pytest.mark.skipif(api_key is None,
+                    reason="Set api_key variable to your OpenAI API key from https://beta.openai.com/account/api-keys")
+def test_understand_message_using_gpt():
+    # Act
+    response = understand("When did I last dine at Subway?", api_key=api_key)
+
+    # Assert
+    assert len(response) > 0
+    assert "remember(\"ledger\", " in response

tests/test_image_search.py

@@ -1,3 +1,6 @@
+# External Packages
+import pytest
+
 # Internal Packages
 from src.main import model
 from src.search_type import image_search
@@ -17,6 +20,7 @@ def test_image_search_setup(search_config):
 
 
 # ----------------------------------------------------------------------------------------------------
+@pytest.mark.skip(reason="results inconsistent currently")
 def test_image_search(search_config):
     # Arrange
     model.image_search = image_search.setup(search_config.image, regenerate=False)