lzy/sci-gui-agent-benchmark
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Save local changes before pulling

Browse Source
This commit is contained in:
kingyang0
2026-03-26 10:50:07 +08:00
parent a38d2faec3
commit 04089fa218
2 changed files with 29 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
evaluation_examples/test_final.json
View File

@@ -23,5 +23,28 @@
"VESTA_Manual_task9",
"VESTA_Manual_task10",
"VESTA_Manual_task11"
],
"origin": [
"Origin_User_Guide_2025b_E_task2",
"Origin_User_Guide_2025b_E_task3",
"Origin_User_Guide_2025b_E_task4",
"Origin_User_Guide_2025b_E_task5",
"Origin_User_Guide_2025b_E_task8",
"Origin_User_Guide_2025b_E_task9",
"Origin_User_Guide_2025b_E_task11",
"Origin_User_Guide_2025b_E_task12"
],
"avogadro": [
"building-metal-complexes_task1",
"building-metal-complexes_task3",
"building-metal-complexes_task7",
"building-organic-molecules_task1",
"building-organic-molecules_task3",
"building-organic-molecules_task4",
"building-organic-molecules_task5",
"building-organic-molecules_task9",
"naming-a-molecule_task1",
"using-qtaim-and-wfn_task2",
"viewing-electrostatic-potential_task1"
]
}

12
run_proxmox.sh
View File

@@ -12,7 +12,7 @@ export PROXMOX_VM_IP="10.10.17.10"
# ---------- LLM API 配置 ----------
# OpenAI 兼容代理（同时用于 Agent 模型和 Eval 模型）
export OPENAI_API_KEY="sk-5zk3CL73E2DsNyMn5a6dA357B6214eEd9240A674Ec0555Be"
export OPENAI_API_KEY="sk-EQGuvk0rS7EG4Cu22cF6D5Cc3a324c88B2E2D432Bc59Bb17" # ⚠️ 请替换为你的实际 API Key
export OPENAI_BASE_URL="https://vip.apiyi.com/v1"
# ---------- 评测参数（按需修改） ----------
@@ -26,13 +26,13 @@ TEMPERATURE=0.5 # 生成温度（越低越稳定
TOP_P=0.9 # nucleus sampling
MAX_TOKENS=16384 # 模型最大输出 token 数
MAX_TRAJECTORY_LENGTH=3 # 历史轨迹保留长度
OBSERVATION_TYPE="screenshot_a11y_tree" # 观测类型
OBSERVATION_TYPE="screenshot" # 观测类型
ACTION_SPACE="pyautogui" # 动作空间
SCREEN_WIDTH=1920 # 屏幕宽度
SCREEN_HEIGHT=1080 # 屏幕高度
RESULT_DIR="/Volumes/Castor/课题/results" # 结果输出目录
TEST_META="evaluation_examples/test_final.json" # 评测任务列表
DOMAIN="vesta" # 评测领域
RESULT_DIR="/mnt/d/work/result" # 结果输出目录
TEST_META="/mnt/d/work/sci-gui-agent-benchmark/evaluation_examples/test_final.json" # 评测任务列表
DOMAIN="origin" # 评测领域
SNAPSHOT_NAME="snapshot" # 快照名称（需提前创建）
INJECT_STEPS=false # 是否注入教程步骤到 Agent prompt（baseline 不注入）
@@ -101,7 +101,7 @@ else
INJECT_STEPS_FLAG="--no_inject_steps"
fi
python run.py \
python3 run.py \
--provider_name "${PROVIDER}" \
--path_to_vm "${VM_ID}" \
--observation_type "${OBSERVATION_TYPE}" \