Merge branch 'main' into feature/lambda-function

src/.env-sample

@@ -7,6 +7,9 @@ FLASK_DEBUG=True                   # True or False
 # OpenAI. Mandatory. Enables language modeling.
 OPENAI_API_KEY=                     # OpenAI API key 
 
+# Temperature configuration for OpenAI. Optional. Default is 0.
+TEMPERATURE=                        # Only applies to the legacy task agent
+
 # Serper.dev. Optional. Enables Google web search capability
 # SERPER_API_KEY=                   # Serper.dev API key
 

src/apps/slackapp/pyproject.toml

@@ -12,6 +12,7 @@ flask-cors = "^4.0.0"
 flask = "^2.3.3"
 loguru = "^0.7.0"
 sherpa-ai = {path = "../..", develop = true}
+hydra-core = "^1.3.2"
 
 [tool.poetry.scripts]
 sherpa_slack = 'slackapp.bolt_app:main'

src/apps/slackapp/slackapp/bolt_app.py

@@ -4,14 +4,16 @@
 ##############################################
 
 import time
-from typing import Dict, List
+from typing import Dict, List, Optional
 
 from flask import Flask, request
 from langchain.schema import AIMessage, BaseMessage, HumanMessage
 from loguru import logger
+from omegaconf import OmegaConf
 from slack_bolt import App
 from slack_bolt.adapter.flask import SlackRequestHandler
 from slackapp.routes.whitelist import whitelist_blueprint
+from slackapp.utils import get_qa_agent_from_config_file
 
 import sherpa_ai.config as cfg
 from sherpa_ai.agents import QAAgent
@@ -92,7 +94,9 @@ def get_response(
     previous_messages: List[BaseMessage],
     verbose_logger: BaseVerboseLogger,
     bot_info: Dict[str, str],
-    llm: SherpaChatOpenAI = None,
+    llm=None,
+    team_id: Optional[str] = None,
+    user_id: Optional[str] = None,
 ) -> str:
     """
     Get response from the task agent for the question
@@ -103,6 +107,8 @@ def get_response(
         verbose_logger (BaseVerboseLogger): verbose logger to be used
         bot_info (Dict[str, str]): information of the Slack bot
         llm (SherpaChatOpenAI, optional): LLM to be used. Defaults to None.
+        team_id (str, optional): team id of the Slack workspace. Defaults to "".
+        user_id (str, optional): user id of the Slack user. Defaults to "".
 
     Returns:
         str: response from the task agent
@@ -120,6 +126,13 @@ def get_response(
     tools = get_tools(memory, agent_config)
 
     if agent_config.use_task_agent:
+        llm = SherpaChatOpenAI(
+            openai_api_key=cfg.OPENAI_API_KEY,
+            user_id=user_id,
+            team_id=team_id,
+            temperature=cfg.TEMPERATURE,
+        )
+
         verbose_logger.log("⚠️🤖 Use task agent (obsolete)...")
         task_agent = TaskAgent.from_llm_and_tools(
             ai_name=ai_name,
@@ -135,21 +148,11 @@ def get_response(
 
         response = error_handler.run_with_error_handling(task_agent.run, task=question)
     else:
-        memory = SharedMemory(objective="Answer the question")
-
+        agent = get_qa_agent_from_config_file("conf/config.yaml", team_id, user_id, llm)
         for message in previous_messages:
-            memory.add(EventType.result, message.type, message.content)
-        memory.add(EventType.task, "human", question)
-
-        agent = QAAgent(
-            llm=llm,
-            name=ai_name,
-            num_runs=1,
-            shared_memory=memory,
-            agent_config=agent_config,
-            require_meta=True,
-            verbose_logger=verbose_logger,
-        )
+            agent.shared_memory.add(EventType.result, message.type, message.content)
+        agent.shared_memory.add(EventType.task, "human", question)
+        agent.verbose_logger = verbose_logger
 
         error_handler = AgentErrorHandler()
         response = error_handler.run_with_error_handling(agent.run)
@@ -242,20 +245,13 @@ def event_test(client, say, event):
             )
             question = reconstructor.reconstruct_prompt()
 
-        llm = SherpaChatOpenAI(
-            openai_api_key=cfg.OPENAI_API_KEY,
-            user_id=user_id,
-            team_id=team_id,
-            verbose_logger = slack_verbose_logger,
-            temperature=cfg.TEMPRATURE,
-        )
-
         results = get_response(
             question,
             previous_messages,
             verbose_logger=slack_verbose_logger,
             bot_info=bot,
-            llm=llm,
+            team_id=team_id,
+            user_id=user_id,
         )
 
         say(results, thread_ts=thread_ts)

src/apps/slackapp/slackapp/utils.py

@@ -0,0 +1,33 @@
+from typing import Optional
+
+from hydra.utils import instantiate
+from langchain.base_language import BaseLanguageModel
+from omegaconf import OmegaConf
+
+from sherpa_ai.agents.qa_agent import QAAgent
+from sherpa_ai.config.task_config import AgentConfig
+
+
+def get_qa_agent_from_config_file(
+    config_path: str,
+    user_id: Optional[str] = None,
+    team_id: Optional[str] = None,
+    llm: Optional[BaseLanguageModel] = None,
+) -> QAAgent:
+    config = OmegaConf.load(config_path)
+
+    agent_config: AgentConfig = instantiate(config.agent_config)
+    if user_id is not None:
+        config["user_id"] = user_id
+
+    if team_id is not None:
+        config["team_id"] = team_id
+
+    if llm is None:
+        qa_agent: QAAgent = instantiate(config.qa_agent, agent_config=agent_config)
+    else:
+        qa_agent: QAAgent = instantiate(
+            config.qa_agent, agent_config=agent_config, llm=llm
+        )
+
+    return qa_agent

src/apps/slackapp/tests/data/test_get_agent.yaml

@@ -0,0 +1,51 @@
+shared_memory:
+  _target_: sherpa_ai.memory.shared_memory.SharedMemory
+  objective: Answer the question
+
+user_id: none
+team_id: none
+
+llm:
+  _target_: sherpa_ai.models.sherpa_base_chat_model.SherpaChatOpenAI
+  model_name: gpt-4
+  temperature: 0.7
+  user_id: ${user_id}
+  team_id: ${team_id}
+
+agent_config:
+  _target_: sherpa_ai.config.task_config.AgentConfig
+
+citation_validation:
+  _target_: sherpa_ai.output_parsers.citation_validation.CitationValidation
+  sequence_threshold: 0.8
+  jaccard_threshold: 0.7
+  token_overlap: 0.6
+
+arxiv_search:
+  _target_: sherpa_ai.actions.arxiv_search.ArxivSearch
+  role_description: Act as a question answering agent
+  task: Question answering
+  llm: ${llm}
+  max_results: 3
+
+google_search:
+  _target_: sherpa_ai.actions.GoogleSearch
+  role_description: Act as a question answering agent
+  task: Question answering
+  llm: ${llm}
+  include_metadata: true
+  config: ${agent_config}
+
+qa_agent:
+  _target_: sherpa_ai.agents.qa_agent.QAAgent
+  llm: ${llm}
+  shared_memory: ${shared_memory}
+  name: QA Sherpa
+  description: Act as a question answering agent
+  agent_config: ${agent_config}
+  num_runs: 1
+  actions:
+    - ${arxiv_search}
+    - ${google_search}
+  validations:
+    - ${citation_validation}

src/apps/slackapp/tests/test_get_agent.py

@@ -0,0 +1,20 @@
+from slackapp.utils import get_qa_agent_from_config_file
+
+from sherpa_ai.actions import ArxivSearch, GoogleSearch
+from sherpa_ai.agents.qa_agent import QAAgent
+from sherpa_ai.output_parsers.citation_validation import CitationValidation
+from sherpa_ai.test_utils.data import get_test_data_file_path
+
+
+def test_get_agent(get_test_data_file_path):  # noqa: F811
+    config_filename = get_test_data_file_path(__file__, "test_get_agent.yaml")
+    agent = get_qa_agent_from_config_file(config_filename)
+
+    assert agent is not None
+    assert type(agent) is QAAgent
+
+    assert len(agent.actions) == 2
+    assert type(agent.actions[0]) is ArxivSearch
+    assert type(agent.actions[1]) is GoogleSearch
+    assert len(agent.validations) == 1
+    assert type(agent.validations[0]) is CitationValidation

src/conf/config.yaml

@@ -0,0 +1,55 @@
+shared_memory:
+  _target_: sherpa_ai.memory.shared_memory.SharedMemory
+  objective: Answer the question
+
+user_id: none
+team_id: none
+
+llm:
+  _target_: sherpa_ai.models.sherpa_base_chat_model.SherpaChatOpenAI
+  model_name: gpt-3.5-turbo
+  temperature: 0
+  user_id: ${user_id}
+  team_id: ${team_id}
+
+agent_config:
+  _target_: sherpa_ai.config.task_config.AgentConfig
+
+citation_validation:
+  _target_: sherpa_ai.output_parsers.citation_validation.CitationValidation
+  sequence_threshold: 0.5
+  jaccard_threshold: 0.5
+  token_overlap: 0.5
+
+number_validation:
+  _target_: sherpa_ai.output_parsers.number_validation.NumberValidation
+
+arxiv_search:
+  _target_: sherpa_ai.actions.arxiv_search.ArxivSearch
+  role_description: Act as a question answering agent
+  task: Question answering
+  llm: ${llm}
+  max_results: 3
+
+google_search:
+  _target_: sherpa_ai.actions.GoogleSearch
+  role_description: Act as a question answering agent
+  task: Question answering
+  llm: ${llm}
+  include_metadata: true
+  config: ${agent_config}
+
+qa_agent:
+  _target_: sherpa_ai.agents.qa_agent.QAAgent
+  llm: ${llm}
+  shared_memory: ${shared_memory}
+  name: QA Sherpa
+  description: Act as a question answering agent
+  agent_config: ${agent_config}
+  num_runs: 1
+  validation_steps: 1
+  actions:
+    - ${google_search}
+  validations:
+    - ${number_validation}
+    - ${citation_validation}