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

Update AWS AMI ID, enhance directory creation logic in file upload, modify osworld service configuration, and refine JSON evaluation examples for improved clarity and functionality.

Browse Source
This commit is contained in:
yuanmengqi
2025-07-03 16:57:41 +00:00
parent 7b2120c843
commit a651b04e49
5 changed files with 55 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
desktop_env/providers/aws/manager.py
View File

@@ -36,7 +36,7 @@ DEFAULT_REGION = "us-east-1"
# todo: Add doc for the configuration of image, security group and network interface # todo: Add doc for the configuration of image, security group and network interface
# todo: public the AMI images # todo: public the AMI images
IMAGE_ID_MAP = { IMAGE_ID_MAP = {
"us-east-1": "ami-0cae20d2680c939d4", "us-east-1": "ami-0a6a55a75d04c8888",
"ap-east-1": "ami-0c092a5b8be4116f5", "ap-east-1": "ami-0c092a5b8be4116f5",
} }

4
desktop_env/server/main.py
View File

@@ -1161,7 +1161,9 @@ def upload_file():
try: try:
# Ensure target directory exists # Ensure target directory exists
os.makedirs(os.path.dirname(file_path), exist_ok=True) target_dir = os.path.dirname(file_path)
if target_dir: # Only create directory if it's not empty
os.makedirs(target_dir, exist_ok=True)
# Save file and get size for verification # Save file and get size for verification
file.save(file_path) file.save(file_path)

13
desktop_env/server/osworld_server.service
View File

@@ -1,16 +1,19 @@
[Unit] [Unit]
Description=OSWorld Server Description=osworld service
StartLimitIntervalSec=60 StartLimitIntervalSec=60
StartLimitBurst=4 StartLimitBurst=4
After=network.target auditd.service After=network.target auditd.service
[Service] [Service]
ExecStart=/usr/bin/python3 /home/user/main.py Type=simple
ExecStart=/usr/bin/python /home/user/server/main.py
User=user User=user
WorkingDirectory=/home/user WorkingDirectory=/home/user
Environment="DISPLAY=:0"
Environment="DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/1000/bus"
Environment="XDG_RUNTIME_DIR=/run/user/1000"
Restart=on-failure Restart=on-failure
RestartSec=1 RestartSec=5s
Environment="DISPLAY=:1"
[Install] [Install]
WantedBy=graphical.target WantedBy=multi-user.target

2
evaluation_examples/examples/os/37887e8c-da15-4192-923c-08fa390a176d.json
View File

@@ -57,7 +57,7 @@
"parameters": { "parameters": {
"files": [ "files": [
{ {
"url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/os/37887e8c-da15-4192-923c-08fa390a176d/eval.sh", "url": "https://huggingface.co/datasets/xlangai/ubuntu_osworld_file_cache/resolve/main/os/37887e8c-da15-4192-923c-08fa390a176d/eval_20250703.sh",
"path": "eval.sh" "path": "eval.sh"
} }
] ]

60
evaluation_examples/examples/os/3ce045a0-877b-42aa-8d2c-b4a863336ab8.json
View File

@@ -20,25 +20,49 @@
"os" "os"
], ],
"evaluator": { "evaluator": {
"func": "check_include_exclude", "func": [
"result": { "check_include_exclude",
"type": "vm_command_line", "check_include_exclude"
"command": "LARGE_TEXT=$(gsettings get org.gnome.desktop.interface text-scaling-factor 2>/dev/null | grep -E '1\\.[2-9]|[2-9]\\.' | wc -l) && ZOOM_ENABLED=$(gsettings get org.gnome.desktop.a11y.applications screen-magnifier-enabled 2>/dev/null | grep -c true) && ZOOM_FACTOR=$(gsettings get org.gnome.desktop.a11y.magnifier mag-factor 2>/dev/null | grep -E '[2-9]\\.|1\\.[1-9]' | wc -l) && if [ \"$LARGE_TEXT\" -gt 0 ]; then echo \"Large text enabled\"; elif [ \"$ZOOM_ENABLED\" -gt 0 ] && [ \"$ZOOM_FACTOR\" -gt 0 ]; then echo \"Zoom enabled with magnification\"; else echo \"No enlargement enabled\"; fi", ],
"shell": true "conj": "or",
}, "result": [
"expected": { {
"type": "rule", "type": "vm_command_line",
"rules": { "command": "TEXT_SCALE=$(timeout 10 gsettings get org.gnome.desktop.interface text-scaling-factor 2>/dev/null | grep -oE '[0-9]+\\.[0-9]+' | head -1); if [ -n \"$TEXT_SCALE\" ] && [ \"$(echo \"$TEXT_SCALE >= 1.2\" | bc -l 2>/dev/null || echo 0)\" = \"1\" ]; then echo \"Large text enabled (factor: $TEXT_SCALE)\"; else echo \"No text scaling enabled\"; fi",
"include": [ "shell": true
"Large text enabled", },
"Zoom enabled with magnification" {
], "type": "vm_command_line",
"exclude": [ "command": "ZOOM_ENABLED=$(timeout 10 gsettings get org.gnome.desktop.a11y.applications screen-magnifier-enabled 2>/dev/null | grep -c true); ZOOM_FACTOR=$(timeout 10 gsettings get org.gnome.desktop.a11y.magnifier mag-factor 2>/dev/null | grep -oE '[0-9]+\\.[0-9]+' | head -1); if [ \"$ZOOM_ENABLED\" -gt 0 ] && [ -n \"$ZOOM_FACTOR\" ] && [ \"$(echo \"$ZOOM_FACTOR > 1.1\" | bc -l 2>/dev/null || echo 0)\" = \"1\" ]; then echo \"Zoom enabled with magnification (factor: $ZOOM_FACTOR)\"; else echo \"No zoom magnification enabled\"; fi",
"No enlargement enabled", "shell": true
"error"
]
} }
} ],
"expected": [
{
"type": "rule",
"rules": {
"include": [
"Large text enabled"
],
"exclude": [
"No text scaling enabled",
"error"
]
}
},
{
"type": "rule",
"rules": {
"include": [
"Zoom enabled with magnification"
],
"exclude": [
"No zoom magnification enabled",
"error"
]
}
}
]
}, },
"proxy": false "proxy": false
} }