Spaces:
Running
Running
openhands
commited on
Commit
·
b42a4fe
1
Parent(s):
b16f7da
Update fallback category mappings: place SWE-Bench Multimodal under 'Frontend Development' and Swt-Bench under 'Test Generation'.\n\nCo-authored-by: openhands <[email protected]>
Browse files- simple_data_loader.py +2 -2
simple_data_loader.py
CHANGED
|
@@ -64,10 +64,10 @@ class SimpleLeaderboardViewer:
|
|
| 64 |
print("[DATA_LOADER] No agenteval.json found, using fallback category mappings")
|
| 65 |
fallback_mappings = {
|
| 66 |
'swe-bench': ['Bug Fixing'],
|
| 67 |
-
'swe-bench-multimodal': ['
|
| 68 |
'commit0': ['App Creation'],
|
| 69 |
'multi-swe-bench': ['Bug Fixing'],
|
| 70 |
-
'swt-bench': ['
|
| 71 |
'gaia': ['Information Gathering'],
|
| 72 |
}
|
| 73 |
for benchmark, categories in fallback_mappings.items():
|
|
|
|
| 64 |
print("[DATA_LOADER] No agenteval.json found, using fallback category mappings")
|
| 65 |
fallback_mappings = {
|
| 66 |
'swe-bench': ['Bug Fixing'],
|
| 67 |
+
'swe-bench-multimodal': ['Frontend Development'],
|
| 68 |
'commit0': ['App Creation'],
|
| 69 |
'multi-swe-bench': ['Bug Fixing'],
|
| 70 |
+
'swt-bench': ['Test Generation'],
|
| 71 |
'gaia': ['Information Gathering'],
|
| 72 |
}
|
| 73 |
for benchmark, categories in fallback_mappings.items():
|