Added new benchmarks
Browse files
app.py
CHANGED
@@ -17,7 +17,7 @@ import re
|
|
17 |
import html
|
18 |
from typing import Dict, Any
|
19 |
|
20 |
-
BENCHMARKS = ["WebArena", "WorkArena-L1", "WorkArena-L2", "WorkArena-L3", "MiniWoB",]
|
21 |
|
22 |
def sanitize_agent_name(agent_name):
|
23 |
# Only allow alphanumeric chars, hyphen, underscore
|
|
|
17 |
import html
|
18 |
from typing import Dict, Any
|
19 |
|
20 |
+
BENCHMARKS = ["WebArena", "WorkArena-L1", "WorkArena-L2", "WorkArena-L3", "MiniWoB", "WebLINX", "AssistantBench"]
|
21 |
|
22 |
def sanitize_agent_name(agent_name):
|
23 |
# Only allow alphanumeric chars, hyphen, underscore
|