lzy/sci-gui-agent-benchmark
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update claude endpoint

Browse Source
This commit is contained in:
Timothyxxx
2024-03-18 14:59:02 +08:00
parent d74ab1a44e
commit c1c7ac298f
2 changed files with 10 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
lib_run_single.py
View File

@@ -62,7 +62,6 @@ def run_single_example(agent, env, example, max_steps, instruction, args, exampl
logger.info("The episode is done.") logger.info("The episode is done.")
break break
step_idx += 1 step_idx += 1
# wandb.log({"str_trajectory": str_table})
run.log({"str_trajectory": str_table}) run.log({"str_trajectory": str_table})
result = env.evaluate() result = env.evaluate()
logger.info("Result: %.2f", result) logger.info("Result: %.2f", result)
@@ -71,4 +70,3 @@ def run_single_example(agent, env, example, max_steps, instruction, args, exampl
f.write(f"{result}\n") f.write(f"{result}\n")
env.controller.end_recording(os.path.join(example_result_dir, "recording.mp4")) env.controller.end_recording(os.path.join(example_result_dir, "recording.mp4"))
run.log({"Result": result}) run.log({"Result": result})
# wandb.log({"Result": result})

15
run.py
View File

@@ -52,8 +52,7 @@ logger = logging.getLogger("desktopenv.experiment")
# wandb config # wandb config
### set your wandb api key here ### set your wandb api key here
os.environ["WANDB_API_KEY"] = "" wandb.login(key=os.get("WANDB_API_KEY", None))
wandb.login(key=os.environ["WANDB_API_KEY"])
def config() -> argparse.Namespace: def config() -> argparse.Namespace:
@@ -279,10 +278,10 @@ if __name__ == '__main__':
left_info += f"{domain}: {len(test_file_list[domain])}\n" left_info += f"{domain}: {len(test_file_list[domain])}\n"
logger.info(f"Left tasks:\n{left_info}") logger.info(f"Left tasks:\n{left_info}")
# get_result(args.action_space, get_result(args.action_space,
# args.model, args.model,
# args.observation_type, args.observation_type,
# args.result_dir, args.result_dir,
# test_all_meta test_all_meta
# ) )
test(args, test_file_list) test(args, test_file_list)