Python: Add Agent Framework Lab Lightning package with RL training examples (#937)

* add math agent

* .

* update

* update debug mode

* add tau2 training

* .

* .

* .

* .

* add tests

* .

* revert observability

* update readme

* fix task serialization issue

* fix exception

* add inline docs

* update readme

* update pyproject toml

* minor fix

* update and use git lfs

* update

* update ignore file to use lab specific

* fix type

* update depedency

---------

Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>

python/packages/lab/tau2/agent_framework_lab_tau2/__init__.py

@@ -5,7 +5,7 @@
 import importlib.metadata
 
 from ._tau2_utils import patch_env_set_state, unpatch_env_set_state
-from .runner import TaskRunner
+from .runner import ASSISTANT_AGENT_ID, ORCHESTRATOR_ID, USER_SIMULATOR_ID, TaskRunner
 
 try:
     __version__ = importlib.metadata.version(__name__)
@@ -13,6 +13,9 @@ except importlib.metadata.PackageNotFoundError:
     __version__ = "0.0.0"  # Fallback for development mode
 
 __all__ = [
+    "ASSISTANT_AGENT_ID",
+    "ORCHESTRATOR_ID",
+    "USER_SIMULATOR_ID",
     "TaskRunner",
     "patch_env_set_state",
     "unpatch_env_set_state",

python/packages/lab/tau2/agent_framework_lab_tau2/runner.py

@@ -32,6 +32,8 @@ from ._message_utils import flip_messages, log_messages
 from ._sliding_window import SlidingWindowChatMessageStore
 from ._tau2_utils import convert_agent_framework_messages_to_tau2_messages, convert_tau2_tool_to_ai_function
 
+__all__ = ["ASSISTANT_AGENT_ID", "ORCHESTRATOR_ID", "USER_SIMULATOR_ID", "TaskRunner"]
+
 # Agent instructions matching tau2's LLMAgent
 ASSISTANT_AGENT_INSTRUCTION = """
 You are a customer service agent that helps the user according to the <policy> provided below.
@@ -413,5 +415,8 @@ class TaskRunner:
             domain="airline",
         )
 
-        logger.info(f"Evaluation completed - Reward: {self.full_reward_info.reward}, Info: {self.full_reward_info}")
-        return self.full_reward_info.reward  # type: ignore[no-any-return]
+        logger.info(
+            f"Evaluation completed - Reward: {self.full_reward_info.reward if self.full_reward_info else None}, "
+            f"Info: {self.full_reward_info}"
+        )
+        return self.full_reward_info.reward if self.full_reward_info else 0.0