update wandb settings

run.py

@@ -148,7 +148,7 @@ def test(
 
     for domain in tqdm(test_all_meta, desc="Domain"):
         for example_id in tqdm(test_all_meta[domain], desc="Example", leave=False):
-            wandb.init(project=f"OSworld-{args.action_space}-{args.observation_type}-{args.model}", group=f"{domain}", 
+            run = wandb.init(project=f"OSworld-{args.action_space}-{args.observation_type}-{args.model}", group=f"{domain}", 
                     name=f"{example_id}")
             # example setting
             config_file = os.path.join(args.test_config_base_dir, f"examples/{domain}/{example_id}.json")
@@ -164,7 +164,7 @@ def test(
             # wandb each example config settings
             cfg_args["instruction"] = instruction
             cfg_args["start_time"] = datetime.datetime.now().strftime("%Y:%m:%d-%H:%M:%S")
-            wandb.config.update(cfg_args)   
+            run.config.update(cfg_args)   
 
             example_result_dir = os.path.join(
                 args.result_dir,
@@ -178,7 +178,7 @@ def test(
             # example start running
             try:
                 lib_run_single.run_single_example(agent, env, example, max_steps, instruction, args, example_result_dir,
-                                                  scores)
+                                                  scores, run)
             except Exception as e:
                 logger.error(f"Exception in {domain}/{example_id}: {e}")
                 wandb.log({"Exception": wandb.Table(data=[[f"Exception in {domain}/{example_id}: {e}"]], columns=["Error"])})