Validation list

tests/validation/ens_check.py

@@ -0,0 +1,7 @@
+import logging
+from protollm_tools.llm-agents-api.protollm_agents.entrypoint import Entrypoint
+logging.basicConfig(level=logging.DEBUG, format="%(asctime)s - %(name)s - %(levelname)s - %(filename)s:%(lineno)d - %(message)s", handlers=[logging.StreamHandler()], )
+logger = logging.getLogger(name)
+if __name__ == "__main__":
+    epoint = Entrypoint(config_path="./examples/admin-config.yml")
+    epoint.run()

tests/validation/rag_check.py

@@ -0,0 +1,15 @@
+import uuid
+from protollm_tools.sdk.protollm_sdk.jobs.utility import construct_job_context
+from protollm_tools.sdk.protollm_sdk.utils.reddis import get_reddis_wrapper, load_result
+from protollm.rags.jobs import RAGJob
+
+# Шаг 1. Инициализация уникального номера идентификации
+job_id = str(uuid.uuid4())
+# Шаг 2.  Инициализация переменных доступа к БД и SDK
+job_name = "fast_validation"
+ctx = construct_job_context(job_name)
+# Шаг 3. Запуск поиска релевантных документов
+RAGJob().run(job_id, ctx, user_prompt='Какой бывает арматура железобетонных конструкций?', use_advanced_rag=False)
+# Шаг 4. Получение ответа модели из базы данных.
+rd = get_reddis_wrapper()
+result = load_result(rd, job_id, job_name)

tests/validation/repro_check.py

@@ -0,0 +1,134 @@
+from langchain.agents import (
+    create_structured_chat_agent,
+    AgentExecutor,
+)
+from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
+from langchain_core.prompts import (
+    SystemMessagePromptTemplate,
+    HumanMessagePromptTemplate,
+)
+from protollm_tools.sdk.protollm_sdk.jobs.job import Job
+from protollm_tools.sdk.protollm_sdk.jobs.job_context import JobContext
+
+from protollm_tools.llm_agents_api.models import StairsLLMAgentResult
+from protollm_tools.llm_agents_api.config import CUSTOM_USER_MESSAGE, CUSTOM_SYSTEM_MESSAGE
+from protollm_tools.llm_agents_api.parse_result import parse_intermediate_steps
+from protollm_tools.llm_agents_api.tools import (
+    query_database_rag,
+    get_time,
+    get_resource,
+    restore_works_edges,
+    start_schedule,
+    extract_scheduling_params
+)
+from protollm.agents.llama31_agents.llama31_agent import Llama31ChatModel
+
+class StairsLLMAgentJob(Job):
+    """
+    Job class representing the main execution of the Stairs LLM Agent.
+
+    Attributes:
+        tools (list): List of tools available to the agent.
+        prompt (ChatPromptTemplate): Chat prompt template used to structure the conversation.
+        llm (Llama31ChatModel): Language model instance used by the agent for generating responses.
+    """
+
+    def __init__(self):
+        """Initializes the StairsLLMAgentJob with tools, prompt, and language model."""
+        super().__init__()
+        self.tools = [
+            query_database_rag,
+            get_time, get_resource,
+            restore_works_edges,
+            start_schedule,
+            extract_scheduling_params
+        ]
+        self.prompt = ChatPromptTemplate.from_messages(
+            [
+                SystemMessagePromptTemplate.from_template(
+                    CUSTOM_SYSTEM_MESSAGE,
+                    input_variables=["tools", "tool_names"],
+                ),
+                MessagesPlaceholder(variable_name="chat_history", optional=True),
+                HumanMessagePromptTemplate.from_template(
+                    CUSTOM_USER_MESSAGE,
+                    input_variables=["input", "agent_scratchpad"],
+                ),
+            ]
+        )
+
+        self.llm = Llama31ChatModel()
+
+    def run(self, job_id: str, ctx: JobContext, **kwargs):
+        """Executes the agent job, generating and saving the result.
+
+        Args:
+            job_id (str): Unique identifier for the job execution.
+            ctx (JobContext): Context object for managing the job's lifecycle and data storage.
+            **kwargs: Additional parameters which includes:
+                request (str): The user query string.
+                is_project (bool): Flag indicating if the request is a project.
+                is_scheduling (bool): Flag indicating if the request is for scheduling.
+
+        Returns:
+            None. Saves the job result directly into the provided job context.
+        """
+        request: str = kwargs.get("request", "")
+        is_project: bool or None = kwargs.get("is_project", None)
+        request += "" if is_project is None else f" is_project={is_project}"
+        is_scheduling: bool or None = kwargs.get("is_scheduling", None)
+        request += "" if is_scheduling is None else f" is_scheduling={is_scheduling}"
+
+        agent_request = {"input": request}  # , "context": {"is_project": is_project, "is_scheduling": is_scheduling}}
+
+        agent = create_structured_chat_agent(
+            llm=self.llm,
+            tools=self.tools,
+            prompt=self.prompt,
+            stop_sequence=True,
+        )
+        agent_executor = AgentExecutor.from_agent_and_tools(
+            agent=agent,
+            tools=self.tools,
+            verbose=True,
+            return_intermediate_steps=True,
+            output_keys=["output"],
+        )
+        result = agent_executor.invoke(agent_request)
+
+        parsed_result = parse_intermediate_steps(result)
+        dumped_result = parsed_result.model_dump()
+        ctx.result_storage.save_dict(job_id, dumped_result)
+        print("written to redis: ", parsed_result)
+
+
+import uuid
+
+from protollm_tools.sdk.protollm_sdk.jobs.utility import construct_job_context
+from protollm_tools.sdk.protollm_sdk.utils.reddis import get_reddis_wrapper, load_result
+
+from protollm_tools.llm_agents_api.jobs import StairsLLMAgentJob
+
+def prepare_test_data(request: str, is_project: str or None, is_scheduling: str or None, tools: list[str]) -> dict:
+    return {
+        "request": request,
+        "is_project": is_project,
+        "is_scheduling": is_scheduling,
+        "tools": tools
+    }
+
+SCHEDULE_QUERY = "Запусти планирования для проекта"
+
+SCHEDULE_TEST = prepare_test_data(SCHEDULE_QUERY, True, False, ["start_schedule"])
+
+def main():
+    job_id = str(uuid.uuid4())
+    ctx = construct_job_context("agent")
+    job = StairsLLMAgentJob()
+    request, is_project, is_scheduling, _ = SCHEDULE_TEST
+    job.run(job_id=job_id, ctx=ctx, request=request, is_project=True, is_scheduling=False)
+    rd = get_reddis_wrapper()
+    result = str(load_result(rd, job_id, "agent"))
+
+if __name__ == "__main__":
+    main()