122324154

This commit is contained in:
Timothyxxx
2024-02-02 14:36:53 +08:00
parent 32bcdd0937
commit 068c6f5769
7 changed files with 436 additions and 146 deletions

View File

@@ -123,9 +123,8 @@ def run_one_example(example, agent, max_steps=10, example_trajectory_dir="exp_tr
logger.info("Environment closed.")
def main(example_class, example_id):
def main(example_class, example_id, gpt4_model = "gpt-4-vision-preview"):
action_space = "pyautogui"
gpt4_model = "gpt-4-vision-preview"
gemini_model = "gemini-pro-vision"
logger.info("Running example %s/%s", example_class, example_id)
@@ -137,7 +136,7 @@ def main(example_class, example_id):
example["snapshot"] = "exp_v5"
api_key = os.environ.get("OPENAI_API_KEY")
agent = GPT4v_Agent(api_key=api_key, instruction=example['instruction'], action_space=action_space,
agent = GPT4v_Agent(api_key=api_key, model=gpt4_model, instruction=example['instruction'], action_space=action_space,
exp="screenshot")
#
# api_key = os.environ.get("GENAI_API_KEY")
@@ -168,35 +167,49 @@ if __name__ == '__main__':
"af630914-714e-4a24-a7bb-f9af687d3b91"
]
calc_list = [
# "eb03d19a-b88d-4de4-8a64-ca0ac66f426b",
# "0bf05a7d-b28b-44d2-955a-50b41e24012a",
# "7a4e4bc8-922c-4c84-865c-25ba34136be1",
# "2bd59342-0664-4ccb-ba87-79379096cc08",
# "ecb0df7a-4e8d-4a03-b162-053391d3afaf",
# "7efeb4b1-3d19-4762-b163-63328d66303b",
# "4e6fcf72-daf3-439f-a232-c434ce416af6",
# "6054afcb-5bab-4702-90a0-b259b5d3217c",
# "abed40dc-063f-4598-8ba5-9fe749c0615d",
# "01b269ae-2111-4a07-81fd-3fcd711993b0",
# "8b1ce5f2-59d2-4dcc-b0b0-666a714b9a14",
"0cecd4f3-74de-457b-ba94-29ad6b5dafb6",
"4188d3a4-077d-46b7-9c86-23e1a036f6c1",
"51b11269-2ca8-4b2a-9163-f21758420e78",
"7e429b8d-a3f0-4ed0-9b58-08957d00b127",
"347ef137-7eeb-4c80-a3bb-0951f26a8aff",
"6e99a1ad-07d2-4b66-a1ce-ece6d99c20a5",
"3aaa4e37-dc91-482e-99af-132a612d40f3",
"37608790-6147-45d0-9f20-1137bb35703d",
"f9584479-3d0d-4c79-affa-9ad7afdd8850",
"d681960f-7bc3-4286-9913-a8812ba3261a",
"21df9241-f8d7-4509-b7f1-37e501a823f7",
"1334ca3e-f9e3-4db8-9ca7-b4c653be7d17",
"357ef137-7eeb-4c80-a3bb-0951f26a8aff",
"aa3a8974-2e85-438b-b29e-a64df44deb4b",
"a01fbce3-2793-461f-ab86-43680ccbae25",
"4f07fbe9-70de-4927-a4d5-bb28bc12c52c",
]
"a9f325aa-8c05-4e4f-8341-9e4358565f4f",
"ecb0df7a-4e8d-4a03-b162-053391d3afaf",
"7efeb4b1-3d19-4762-b163-63328d66303b",
"4e6fcf72-daf3-439f-a232-c434ce416af6",
"6054afcb-5bab-4702-90a0-b259b5d3217c",
"abed40dc-063f-4598-8ba5-9fe749c0615d",
"01b269ae-2111-4a07-81fd-3fcd711993b0",
"8b1ce5f2-59d2-4dcc-b0b0-666a714b9a14",
"af2b02f7-acee-4be4-8b66-499fab394915",
"da1d63b8-fa12-417b-ba18-f748e5f770f3",
"636380ea-d5f6-4474-b6ca-b2ed578a20f1",
"5ba77536-05c5-4aae-a9ff-6e298d094c3e",
"4bc4eaf4-ca5e-4db2-8138-8d4e65af7c0b",
"672a1b02-c62f-4ae2-acf0-37f5fb3052b0",
"648fe544-16ba-44af-a587-12ccbe280ea6",
"8985d1e4-5b99-4711-add4-88949ebb2308",
"9e606842-2e27-43bf-b1d1-b43289c9589b",
"fcb6e45b-25c4-4087-9483-03d714f473a9",
"68c0c5b7-96f3-4e87-92a7-6c1b967fd2d2",
"fff629ea-046e-4793-8eec-1a5a15c3eb35",
"5c9a206c-bb00-4fb6-bb46-ee675c187df5",
"e975ae74-79bd-4672-8d1c-dc841a85781d",
"34a6938a-58da-4897-8639-9b90d6db5391",
"b5a22759-b4eb-4bf2-aeed-ad14e8615f19",
"2f9913a1-51ed-4db6-bfe0-7e1c95b3139e",
"2558031e-401d-4579-8e00-3ecf540fb492",
"0cecd4f3-74de-457b-ba94-29ad6b5dafb6",
"4188d3a4-077d-46b7-9c86-23e1a036f6c1",
"51b11269-2ca8-4b2a-9163-f21758420e78",
"7e429b8d-a3f0-4ed0-9b58-08957d00b127",
"347ef137-7eeb-4c80-a3bb-0951f26a8aff",
"6e99a1ad-07d2-4b66-a1ce-ece6d99c20a5",
"3aaa4e37-dc91-482e-99af-132a612d40f3",
"37608790-6147-45d0-9f20-1137bb35703d",
"f9584479-3d0d-4c79-affa-9ad7afdd8850",
"d681960f-7bc3-4286-9913-a8812ba3261a",
"21df9241-f8d7-4509-b7f1-37e501a823f7",
"1334ca3e-f9e3-4db8-9ca7-b4c653be7d17",
"357ef137-7eeb-4c80-a3bb-0951f26a8aff",
"aa3a8974-2e85-438b-b29e-a64df44deb4b",
"a01fbce3-2793-461f-ab86-43680ccbae25",
"4f07fbe9-70de-4927-a4d5-bb28bc12c52c"
]
# for example_id in calc_list:
# main("libreoffice_calc", example_id)
@@ -246,13 +259,13 @@ if __name__ == '__main__':
#
vs_code_list = [
"0ed39f63-6049-43d4-ba4d-5fa2fe04a951",
"53ad5833-3455-407b-bbc6-45b4c79ab8fb",
"eabc805a-bfcf-4460-b250-ac92135819f6",
"982d12a5-beab-424f-8d38-d2a48429e511",
"4e60007a-f5be-4bfc-9723-c39affa0a6d3",
"e2b5e914-ffe1-44d2-8e92-58f8c5d92bb2",
"9439a27b-18ae-42d8-9778-5f68f891805e",
# "0ed39f63-6049-43d4-ba4d-5fa2fe04a951",
# "53ad5833-3455-407b-bbc6-45b4c79ab8fb",
# "eabc805a-bfcf-4460-b250-ac92135819f6",
# "982d12a5-beab-424f-8d38-d2a48429e511",
# "4e60007a-f5be-4bfc-9723-c39affa0a6d3",
# "e2b5e914-ffe1-44d2-8e92-58f8c5d92bb2",
# "9439a27b-18ae-42d8-9778-5f68f891805e",
"ea98c5d7-3cf9-4f9b-8ad3-366b58e0fcae",
"930fdb3b-11a8-46fe-9bac-577332e2640e",
"276cc624-87ea-4f08-ab93-f770e3790175",
@@ -266,28 +279,28 @@ if __name__ == '__main__':
# logger.error("An error occurred while running the example: %s", e)
# continue
multiple_list = [
"f8cfa149-d1c1-4215-8dac-4a0932bad3c2",
"897e3b53-5d4d-444b-85cb-2cdc8a97d903",
"4e9f0faf-2ecc-4ae8-a804-28c9a75d1ddc",
"b52b40a5-ad70-4c53-b5b0-5650a8387052",
"46407397-a7d5-4c6b-92c6-dbe038b1457b",
"2b9493d7-49b8-493a-a71b-56cd1f4d6908",
"51f5801c-18b3-4f25-b0c3-02f85507a078",
"2c9fc0de-3ee7-45e1-a5df-c86206ad78b5",
"510f64c8-9bcc-4be1-8d30-638705850618",
"937087b6-f668-4ba6-9110-60682ee33441",
"ee9a3c83-f437-4879-8918-be5efbb9fac7",
"3680a5ee-6870-426a-a997-eba929a0d25c",
"e135df7c-7687-4ac0-a5f0-76b74438b53e",
"58565672-7bfe-48ab-b828-db349231de6b",
"2fe4b718-3bd7-46ec-bdce-b184f5653624"
]
for example_id in multiple_list:
try:
main("multi_apps", example_id)
except Exception as e:
logger.error("An error occurred while running the example: %s", e)
continue
# multiple_list = [
# "f8cfa149-d1c1-4215-8dac-4a0932bad3c2",
# "897e3b53-5d4d-444b-85cb-2cdc8a97d903",
# "4e9f0faf-2ecc-4ae8-a804-28c9a75d1ddc",
# "b52b40a5-ad70-4c53-b5b0-5650a8387052",
# "46407397-a7d5-4c6b-92c6-dbe038b1457b",
# "2b9493d7-49b8-493a-a71b-56cd1f4d6908",
# "51f5801c-18b3-4f25-b0c3-02f85507a078",
# "2c9fc0de-3ee7-45e1-a5df-c86206ad78b5",
# "510f64c8-9bcc-4be1-8d30-638705850618",
# "937087b6-f668-4ba6-9110-60682ee33441",
# "ee9a3c83-f437-4879-8918-be5efbb9fac7",
# "3680a5ee-6870-426a-a997-eba929a0d25c",
# "e135df7c-7687-4ac0-a5f0-76b74438b53e",
# "58565672-7bfe-48ab-b828-db349231de6b",
# "2fe4b718-3bd7-46ec-bdce-b184f5653624"
# ]
#
# for example_id in multiple_list:
# try:
# main("multi_apps", example_id)
# except Exception as e:
# logger.error("An error occurred while running the example: %s", e)
# continue