File tree Expand file tree Collapse file tree 1 file changed +3
-1
lines changed
tools/submission/submission_checker Expand file tree Collapse file tree 1 file changed +3
-1
lines changed Original file line number Diff line number Diff line change 185185 # TODO: Placeholder for now
186186 "gpt-oss-120b" : ("exact_match" , 83.13 * 0.99 ),
187187 # TODO: Placeholder for now
188- "qwen3-vl-235b-a22b" : ("F1 " , 0.7903 * 0.99 ),
188+ "qwen3-vl-235b-a22b" : ("F1_HIERARCHICAL " , 0.7903 * 0.99 ),
189189 "dlrm-v3" : ("AUC" , 78.663 * 0.99 ), # TODO: Placeholder for now
190190 "yolo-95" : ("mAP" , 53.4 * 0.95 ),
191191 "yolo-99" : ("mAP" , 53.4 * 0.99 ),
11531153 "pointpainting" : 6636 ,
11541154 "yolo-99" : 1525 ,
11551155 "yolo-95" : 1525 ,
1156+ "qwen3-vl-235b-a22b" : 48289
11561157}
11571158
11581159SCENARIO_MAPPING = {
13871388 "mbxp_accuracy" : r".*'mbxp':\s([\d.]+).*" ,
13881389 "exact_match" : r".*'exact_match':\s([\d.]+).*" ,
13891390 "vbench_score" : r".*'vbench_score':\s([\d.]+).*" ,
1391+ "F1_HIERARCHICAL" : r'\{.*"f1":\s*([\d\.]+).*\}' ,
13901392}
13911393
13921394SYSTEM_DESC_REQUIRED_FIELDS = [
You can’t perform that action at this time.
0 commit comments