add app react agent

.gitignore

@@ -25,4 +25,5 @@ step_experiences/*
 build/*
 *.egg-info/*
 cookbook/simple_demo/test_appworld/data/*
-cookbook/simple_demo/test_appworld/experiments/*
+cookbook/simple_demo/test_appworld/experiments/*
+cookbook/simple_demo/test_appworld/exp_result/*

cookbook/simple_demo/test_appworld/appworld_react_agent.py

@@ -1,4 +1,3 @@
-import json
 import os
 
 os.environ["APPWORLD_ROOT"] = "."
@@ -8,6 +7,7 @@ load_dotenv("../../../.env")
 
 import re
 import time
+import json
 
 from appworld import AppWorld, load_task_ids
 from jinja2 import Template
@@ -16,18 +16,21 @@ from openai import OpenAI
 
 from prompt import PROMPT_TEMPLATE
 
-@ray.remote
+
+# @ray.remote
 class AppworldReactAgent:
     """A minimal ReAct Agent for AppWorld tasks."""
 
     def __init__(self,
+                 index: int,
                  task_id: str,
                  experiment_name: str,
                  model_name: str = "qwen3-8b",
                  temperature: float = 0.9,
-                 max_interactions: int = 50,
+                 max_interactions: int = 30,
                  max_response_size: int = 2000):
 
+        self.index: int = index
         self.task_id: str = task_id
         self.experiment_name: str = experiment_name
         self.model_name: str = model_name
@@ -47,7 +50,6 @@ class AppworldReactAgent:
                     model=self.model_name,
                     messages=messages,
                     temperature=self.temperature,
-                    max_tokens=400,
                     extra_body={"enable_thinking": False},
                     seed=0)
 
@@ -109,9 +111,9 @@ class AppworldReactAgent:
                     output = self.next_step(code)
                     self.history.append({"role": "user", "content": output})
 
-                    logger.info(f"task_id={self.task_id} iteration={i} "
-                                f"code=\n{code}\n output=\n{output}\n "
-                                f"score={self.get_reward():.4f}")
+                    logger.info(f"index={self.index} task_id={self.task_id} iteration={i} ")
+                    # f"code=\n{code}\n output=\n{output}\n "
+                    # f"score={self.get_reward():.4f}")
 
                     if self.world.task_completed():
                         break
@@ -127,7 +129,8 @@ class AppworldReactAgent:
                     "uplift_score": uplift_score,
                     "task_history": self.history,
                 }
-
+                # logger.info(f"result={json.dumps(result)}")
+                # p_bar.close()
                 return result
 
         except Exception as e:
@@ -138,7 +141,7 @@ class AppworldReactAgent:
 def main():
     dataset_name = "train"
     task_ids = load_task_ids(dataset_name)
-    agent = AppworldReactAgent(task_id=task_ids[0], experiment_name=f"jinli_{dataset_name}")
+    agent = AppworldReactAgent(index=0, task_id=task_ids[0], experiment_name=f"jinli_{dataset_name}")
     result = agent.execute()
     logger.info(f"result={json.dumps(result)}")
 

cookbook/simple_demo/test_appworld/run_appworld.py

@@ -1,13 +1,15 @@
 import os
 
+import ray
+from ray import logger
+from tqdm import tqdm
+
 os.environ["APPWORLD_ROOT"] = "."
 from dotenv import load_dotenv
 
 load_dotenv("../../../.env")
 
 import json
-import time
-from concurrent.futures import ProcessPoolExecutor
 from pathlib import Path
 
 from appworld import load_task_ids
@@ -15,29 +17,40 @@ from appworld import load_task_ids
 from appworld_react_agent import AppworldReactAgent
 
 
-def run_agent(dataset_name: str, max_workers: int, experiment_suffix: str):
+def run_agent(dataset_name: str, experiment_suffix: str, multi_thread: bool = False):
     experiment_name = dataset_name + "_" + experiment_suffix
     path: Path = Path(f"./exp_result")
     path.mkdir(parents=True, exist_ok=True)
 
     task_ids = load_task_ids(dataset_name)
     result: list = []
-    with ProcessPoolExecutor(max_workers=max_workers) as executor:
-        task_list: list = []
+
+    def dump_file():
+        with open(path / f"{experiment_name}.jsonl", "w") as f:
+            for x in result:
+                f.write(json.dumps(x) + "\n")
+
+    if multi_thread:
+        future_list: list = []
         for index, task_id in enumerate(task_ids):
-            agent = AppworldReactAgent(task_id, experiment_name)
-            task = executor.submit(agent.execute)
-            task_list.append(task)
-            time.sleep(1)
+            actor = AppworldReactAgent.remote(index=index, task_id=task_id, experiment_name=experiment_name)
+            future = actor.execute.remote()
+            future_list.append(future)
+        logger.info("submit complete")
 
-        for task in task_list:
-            result.append(task.result(timeout=600))
+        for future in future_list:
+            result.append(ray.get(future))
+            dump_file()
+
+    else:
+        for index, task_id in enumerate(task_ids):
+            agent = AppworldReactAgent(index=index, task_id=task_id, experiment_name=experiment_name)
+            result.append(agent.execute())
+            dump_file()
 
-    with open(path / f"{experiment_name}.jsonl", "w") as f:
-        for result_item in result:
-            f.write(json.dumps(result_item) + "\n")
 
 
 if __name__ == "__main__":
-    run_agent(dataset_name="train", experiment_suffix="v1", max_workers=1)
-    # run_agent(dataset_name="dev", experiment_suffix="v1", max_workers=1)
+    # ray.init(num_cpus=8)
+    run_agent(dataset_name="train", experiment_suffix="v2")
+    # run_agent(dataset_name="dev", experiment_suffix="v2")