Merge pull request #289 from sirajperson/main

Added local llm functionality by incorporating text-generation-webui

Dockerfile

@@ -14,4 +14,4 @@ COPY entrypoint.sh /entrypoint.sh
 COPY wait-for-it.sh /wait-for-it.sh
 RUN chmod +x /entrypoint.sh /wait-for-it.sh
 
-CMD ["/wait-for-it.sh", "super__postgres:5432","-t","60","--","/entrypoint.sh"]
+CMD ["/wait-for-it.sh", "super__postgres:5432","-t","60","--","/entrypoint.sh"]

DockerfileCelery

@@ -12,4 +12,4 @@ WORKDIR /app
 COPY . .
 COPY config.yaml .
 
-CMD ["celery", "-A", "superagi.worker", "worker", "--loglevel=info"]
+CMD ["celery", "-A", "superagi.worker", "worker", "--loglevel=info"]

cli2.py

@@ -24,10 +24,10 @@ def run_npm_commands(shell=False):
 
 
 def run_server(shell=False,a_name=None,a_description=None,goals=None):
+    tgwui_process = Process(target=subprocess.run, args=(["python", "test.py","--name",a_name,"--description",a_description,"--goals"]+goals,), kwargs={"shell": shell})
     api_process = Process(target=subprocess.run, args=(["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"],), kwargs={"shell": shell})
     celery_process = Process(target=subprocess.run, args=(["celery", "-A", "celery_app", "worker", "--loglevel=info"],), kwargs={"shell": shell})
     ui_process = Process(target=subprocess.run, args=(["python", "test.py","--name",a_name,"--description",a_description,"--goals"]+goals,), kwargs={"shell": shell})
-
     api_process.start()
     celery_process.start()
     ui_process.start()

config_template.yaml

@@ -4,9 +4,17 @@ PINECONE_ENVIRONMENT: YOUR_PINECONE_ENVIRONMENT
 
 OPENAI_API_KEY: YOUR_OPEN_API_KEY
 
-#DATABASE INFO
+# For locally hosted LLMs comment out the next line and uncomment the one after
+# to configure a local llm point your browser to 127.0.0.1:7860 and click on the model tab in text generation web ui.
+OPENAI_API_BASE: https://api.openai.com/v1
+#OPENAI_API_BASE: "http://super__tgwui:5001/v1"
+
+# "gpt-3.5-turbo-0301": 4032, "gpt-4-0314": 8092, "gpt-3.5-turbo": 4032, "gpt-4": 8092, "llama":2048, "mpt-7b-storywriter":45000
+MODEL_NAME: "gpt-3.5-turbo-0301"
 MAX_TOOL_TOKEN_LIMIT: 800
+MAX_MODEL_TOKEN_LIMIT: 4032 # set to 2048 for llama
 
+#DATABASE INFO
 # redis details
 DB_NAME: super_agi_main
 POSTGRES_URL: super__postgres
@@ -48,7 +56,7 @@ ENCRYPTION_KEY: secret
 GOOGLE_API_KEY: YOUR_GOOGLE_API_KEY
 SEARCH_ENGINE_ID: YOUR_SEARCH_ENIGNE_ID
 
-# IF YOU DONT HAVE GOOGLE SERACH KEY, USE THIS
+# IF YOU DONT HAVE GOOGLE SEARCH KEY, USE THIS
 SERP_API_KEY: YOUR_SERP_API_KEY
 
 #ENTER YOUR EMAIL CREDENTIALS TO ACCESS EMAIL TOOL

docker-compose.yaml

@@ -24,7 +24,7 @@ services:
       - super__postgres
     networks:
       - super_network
-
+    
   gui:
     build: ./gui
     ports:
@@ -39,12 +39,12 @@ services:
       - /app/.next
 
   super__redis:
-    image: "redis:latest"
+    image: "docker.io/library/redis:latest"
     networks:
       - super_network
 
   super__postgres:
-    image: "postgres:latest"
+    image: "docker.io/library/postgres:latest"
     environment:
       - POSTGRES_USER=superagi
       - POSTGRES_PASSWORD=password
@@ -61,4 +61,4 @@ networks:
     driver: bridge
 
 volumes:
-  superagi_postgres_data:
+  superagi_postgres_data:

local-llm

@@ -0,0 +1,93 @@
+version: '3.8'
+
+services:
+  backend:
+    volumes:
+      - "./:/app"
+    build: .
+    ports:
+      - "8001:8001"
+    depends_on:
+      - super__tgwui
+      - super__redis
+      - super__postgres
+    networks:
+      - super_network
+
+  celery:
+    volumes:
+      - "./:/app"
+    build:
+      context: .
+      dockerfile: DockerfileCelery
+    depends_on:
+      - super__tgwui
+      - super__redis
+      - super__postgres
+    networks:
+      - super_network
+
+  gui:
+    build: ./gui
+    ports:
+      - "3000:3000"
+    environment:
+      - NEXT_PUBLIC_API_BASE_URL=http://localhost:8001
+    networks:
+      - super_network
+    volumes:
+      - ./gui:/app
+      - /app/node_modules
+      - /app/.next
+
+  super__tgwui:
+    build:
+      context: .
+      dockerfile: ./tgwui/DockerfileTGWUI
+    container_name: super__tgwui
+    environment:
+      - EXTRA_LAUNCH_ARGS="--listen --verbose --extensions openai --threads 4 --n_ctx 1600"
+    ports:
+      - 7860:7860  # Default web port
+      - 5000:5000  # Default API port
+      - 5005:5005  # Default streaming port
+      - 5001:5001  # Default OpenAI API extension port
+    volumes:
+      - ./tgwui/config/loras:/app/loras
+      - ./tgwui/config/models:/app/models
+      - ./tgwui/config/presets:/app/presets
+      - ./tgwui/config/prompts:/app/prompts
+      - ./tgwui/config/softprompts:/app/softprompts
+      - ./tgwui/config/training:/app/training
+    logging:
+      driver:  json-file
+      options:
+        max-file: "3"   # number of files or file count
+        max-size: '10m'
+    networks:
+      - super_network
+
+  super__redis:
+    image: "docker.io/library/redis:latest"
+    networks:
+      - super_network
+
+  super__postgres:
+    image: "docker.io/library/postgres:latest"
+    environment:
+      - POSTGRES_USER=superagi
+      - POSTGRES_PASSWORD=password
+      - POSTGRES_DB=super_agi_main
+    volumes:
+      - superagi_postgres_data:/var/lib/postgresql/data/
+    networks:
+      - super_network
+    ports:
+      - "5432:5432"
+
+networks:
+  super_network:
+    driver: bridge
+
+volumes:
+  superagi_postgres_data:

local-llm-gpu

@@ -0,0 +1,107 @@
+version: '3.8'
+
+services:
+  backend:
+    volumes:
+      - "./:/app"
+    build: .
+    ports:
+      - "8001:8001"
+    depends_on:
+      - super__tgwui
+      - super__redis
+      - super__postgres
+    networks:
+      - super_network
+
+  celery:
+    volumes:
+      - "./:/app"
+    build:
+      context: .
+      dockerfile: DockerfileCelery
+    depends_on:
+      - super__tgwui
+      - super__redis
+      - super__postgres
+    networks:
+      - super_network
+
+  gui:
+    build: ./gui
+    ports:
+      - "3000:3000"
+    environment:
+      - NEXT_PUBLIC_API_BASE_URL=http://localhost:8001
+    networks:
+      - super_network
+    volumes:
+      - ./gui:/app
+      - /app/node_modules
+      - /app/.next
+
+  super__tgwui:
+    build:
+      context: .
+      target: llama-cublas
+      dockerfile: ./tgwui/DockerfileTGWUI
+#      args:
+#        - LCL_SRC_DIR=text-generation-webui  # Developers - see Dockerfile app_base
+    container_name: super__tgwui
+    environment:
+      - EXTRA_LAUNCH_ARGS="--listen --no-mmap --verbose --extensions openai --auto-devices --n_ctx 1600 --gpu-memory 20 20 --n-gpu-layers 128 --threads 8  --model vicuna-13b-cot.ggmlv3.q8_0.bin"
+    ports:
+      - 7860:7860  # Default web port
+      - 5000:5000  # Default API port
+      - 5005:5005  # Default streaming port
+      - 5001:5001  # Default OpenAI API extension port
+    volumes:
+      - ./tgwui/config/loras:/app/loras
+      - ./tgwui/config/models:/app/models
+      - ./tgwui/config/presets:/app/presets
+      - ./tgwui/config/prompts:/app/prompts
+      - ./tgwui/config/softprompts:/app/softprompts
+      - ./tgwui/config/training:/app/training
+      - ./tgwui/config/embeddings:/app/embeddings
+    logging:
+      driver:  json-file
+      options:
+        max-file: "3"   # number of files or file count
+        max-size: '10m'
+    networks:
+      - super_network
+###  Uncomment the following lines to run the container using the host machine's GPU resources
+    deploy:
+        resources:
+          reservations:
+            devices:
+              - driver: nvidia
+#                count: "all"
+                device_ids: ['0', '1'] # must comment the above line if this line is uncommented.
+                capabilities: [gpu]
+
+
+  super__redis:
+    image: "docker.io/library/redis:latest"
+    networks:
+      - super_network
+
+  super__postgres:
+    image: "docker.io/library/postgres:latest"
+    environment:
+      - POSTGRES_USER=superagi
+      - POSTGRES_PASSWORD=password
+      - POSTGRES_DB=super_agi_main
+    volumes:
+      - superagi_postgres_data:/var/lib/postgresql/data/
+    networks:
+      - super_network
+    ports:
+      - "5432:5432"
+
+networks:
+  super_network:
+    driver: bridge
+
+volumes:
+  superagi_postgres_data:

run.sh

@@ -6,6 +6,12 @@ if [ ! -f "config.yaml" ]; then
     exit 1
 fi
 
+if [ ! -f "tgwui/text-generation-webui" ]; then
+    echo "Downloading tgwui src"
+    git clone https://github.com/oobabooga/text-generation-webui
+    mv text-generation-webui tgwui
+fi
+
 # Function to check if virtual environment is activated
 is_venv_activated() {
     [[ -n "$VIRTUAL_ENV" ]]

superagi/agent/super_agi.py

@@ -117,7 +117,7 @@ def execute(self, workflow_step: AgentWorkflowStep):
         agent_execution_id = self.agent_config["agent_execution_id"]
         task_queue = TaskQueue(str(agent_execution_id))
 
-        token_limit = TokenCounter.token_limit(self.llm.get_model())
+        token_limit = TokenCounter.token_limit()
         agent_feeds = self.fetch_agent_feeds(session, self.agent_config["agent_execution_id"], self.agent_config["agent_id"])
         current_calls = 0
         if len(agent_feeds) <= 0:
@@ -283,7 +283,7 @@ def build_agent_prompt(self, prompt: str, task_queue: TaskQueue, max_token_limit
             last_task = response["task"]
             last_task_result = response["response"]
         current_task = task_queue.get_first_task() or ""
-        token_limit = TokenCounter.token_limit(self.llm.get_model()) - max_token_limit
+        token_limit = TokenCounter.token_limit() - max_token_limit
         prompt = AgentPromptBuilder.replace_task_based_variables(prompt, current_task, last_task, last_task_result,
                                                                  pending_tasks, completed_tasks, token_limit)
         return prompt

superagi/helper/json_cleaner.py

@@ -51,7 +51,7 @@ def extract_json_section(cls, input_str: str = ""):
 
     @classmethod
     def remove_escape_sequences(cls, string):
-        return string.encode('utf-8').decode('unicode_escape').encode('raw_unicode_escape').decode('utf-8')
+        return string.encode('utf-8').decode('unicode_escape').encode('raw_unicode_escape')
 
     @classmethod
     def add_quotes_to_property_names(cls, json_string: str) -> str:

superagi/jobs/agent_executor.py

@@ -132,7 +132,7 @@ def execute_next_action(self, agent_execution_id):
                 memory = VectorFactory.get_vector_storage("PineCone", "super-agent-index1",
                                                           OpenAiEmbedding(model_api_key))
         except:
-            print("Unable to setup the pincone connection...")
+            print("Unable to setup the pinecone connection...")
             memory = None
 
         user_tools = session.query(Tool).filter(Tool.id.in_(parsed_config["tools"])).all()

superagi/llms/openai.py

@@ -8,7 +8,7 @@
 
 
 class OpenAi(BaseLlm):
-    def __init__(self, api_key, image_model=None, model="gpt-4", temperature=0.6, max_tokens=4032, top_p=1,
+    def __init__(self, api_key, image_model=None, model="gpt-4", temperature=0.6, max_tokens=get_config("MAX_MODEL_TOKEN_LIMIT"), top_p=1,
                  frequency_penalty=0,
                  presence_penalty=0, number_of_results=1):
         self.model = model
@@ -21,14 +21,15 @@ def __init__(self, api_key, image_model=None, model="gpt-4", temperature=0.6, ma
         self.api_key = api_key
         self.image_model = image_model
         openai.api_key = api_key
+        openai.api_base = get_config("OPENAI_API_BASE", "https://api.openai.com/v1")
 
     def get_model(self):
         return self.model
 
     def get_image_model(self):
         return self.image_model
 
-    def chat_completion(self, messages, max_tokens=4032):
+    def chat_completion(self, messages, max_tokens=get_config("MAX_MODEL_TOKEN_LIMIT")):
         try:
             # openai.api_key = get_config("OPENAI_API_KEY")
             response = openai.ChatCompletion.create(

superagi/tools/email/send_email_attachment.py

@@ -17,11 +17,11 @@ class SendEmailAttachmentInput(BaseModel):
     to: str = Field(..., description="Email Address of the Receiver, default email address is 'example@example.com'")
     subject: str = Field(..., description="Subject of the Email to be sent")
     body: str = Field(..., description="Email Body to be sent")
-    filename: str = Field(..., description="Name of the file to be sent as an Attachement with Email")
+    filename: str = Field(..., description="Name of the file to be sent as an Attachment with Email")
 
 
 class SendEmailAttachmentTool(BaseTool):
-    name: str = "Send Email with Attachement"
+    name: str = "Send Email with Attachment"
     args_schema: Type[BaseModel] = SendEmailAttachmentInput
     description: str = "Send an Email with a file attached to it"
 
@@ -32,9 +32,9 @@ def _execute(self, to: str, subject: str, body: str, filename: str) -> str:
         base_path = base_path + filename
         attachmentpath = base_path
         attachment = os.path.basename(attachmentpath)
-        return self.send_email_with_attachement(to, subject, body, attachmentpath, attachment)
+        return self.send_email_with_attachment(to, subject, body, attachmentpath, attachment)
 
-    def send_email_with_attachement(self, to, subject, body, attachment_path, attachment) -> str:
+    def send_email_with_attachment(self, to, subject, body, attachment_path, attachment) -> str:
         email_sender = get_config('EMAIL_ADDRESS')
         email_password = get_config('EMAIL_PASSWORD')
         if email_sender == "" or email_sender.isspace():