ss
Browse files- src/about.py +5 -1
- src/display/utils.py +10 -2
src/about.py
CHANGED
@@ -14,12 +14,16 @@ class Tasks(Enum):
|
|
14 |
# task_key in the json file, metric_key in the json file, name to display in the leaderboard
|
15 |
CMMMU = Task("CMMMU", "acc", "CMMMU")
|
16 |
MMMU = Task("MMMU", "acc", "MMMU")
|
17 |
-
OCRBench = Task("OCRBench", "acc", "OCRBench")
|
18 |
MMMU_Pro_standard = Task("MMMU_Pro_standard", "acc", "MMMU_Pro_standard")
|
19 |
MMMU_Pro_vision = Task("MMMU_Pro_vision", "acc", "MMMU_Pro_vision")
|
20 |
MathVision = Task("MathVision", "acc", "MathVision")
|
21 |
CII_Bench = Task("CII-Bench", "acc", "CII-Bench")
|
22 |
Blink = Task("Blink", "acc", "Blink")
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
NUM_FEWSHOT = 0 # Change with your few shot
|
25 |
# ---------------------------------------------------
|
|
|
14 |
# task_key in the json file, metric_key in the json file, name to display in the leaderboard
|
15 |
CMMMU = Task("CMMMU", "acc", "CMMMU")
|
16 |
MMMU = Task("MMMU", "acc", "MMMU")
|
|
|
17 |
MMMU_Pro_standard = Task("MMMU_Pro_standard", "acc", "MMMU_Pro_standard")
|
18 |
MMMU_Pro_vision = Task("MMMU_Pro_vision", "acc", "MMMU_Pro_vision")
|
19 |
MathVision = Task("MathVision", "acc", "MathVision")
|
20 |
CII_Bench = Task("CII-Bench", "acc", "CII-Bench")
|
21 |
Blink = Task("Blink", "acc", "Blink")
|
22 |
+
CharXiv = Task("CharXiv", "acc", "CharXiv")
|
23 |
+
MathVerse = Task("MathVerse", "acc", "MathVerse")
|
24 |
+
MmvetV2 = Task("MmvetV2", "acc", "MmvetV2")
|
25 |
+
Ocrlite = Task("Ocrlite", "acc", "Ocrlite")
|
26 |
+
OcrliteZh = Task("OcrliteZh", "acc", "OcrliteZh")
|
27 |
|
28 |
NUM_FEWSHOT = 0 # Change with your few shot
|
29 |
# ---------------------------------------------------
|
src/display/utils.py
CHANGED
@@ -148,10 +148,14 @@ baseline_row = {
|
|
148 |
AutoEvalColumn.MMMU.name: 100,
|
149 |
AutoEvalColumn.MMMU_Pro_standard.name: 100,
|
150 |
AutoEvalColumn.MMMU_Pro_vision.name: 100,
|
151 |
-
AutoEvalColumn.OCRBench.name: 100,
|
152 |
AutoEvalColumn.MathVision.name: 100,
|
153 |
AutoEvalColumn.CII_Bench.name: 100,
|
154 |
AutoEvalColumn.Blink.name: 100,
|
|
|
|
|
|
|
|
|
|
|
155 |
AutoEvalColumn.dummy.name: "baseline",
|
156 |
AutoEvalColumn.model_type.name: "",
|
157 |
AutoEvalColumn.flagged.name: False,
|
@@ -168,10 +172,14 @@ human_baseline_row = {
|
|
168 |
AutoEvalColumn.MMMU.name: 100,
|
169 |
AutoEvalColumn.MMMU_Pro_standard.name: 100,
|
170 |
AutoEvalColumn.MMMU_Pro_vision.name: 100,
|
171 |
-
AutoEvalColumn.OCRBench.name: 100,
|
172 |
AutoEvalColumn.MathVision.name: 100,
|
173 |
AutoEvalColumn.CII_Bench.name: 100,
|
174 |
AutoEvalColumn.Blink.name: 100,
|
|
|
|
|
|
|
|
|
|
|
175 |
AutoEvalColumn.dummy.name: "human_baseline",
|
176 |
AutoEvalColumn.model_type.name: "",
|
177 |
AutoEvalColumn.flagged.name: False,
|
|
|
148 |
AutoEvalColumn.MMMU.name: 100,
|
149 |
AutoEvalColumn.MMMU_Pro_standard.name: 100,
|
150 |
AutoEvalColumn.MMMU_Pro_vision.name: 100,
|
|
|
151 |
AutoEvalColumn.MathVision.name: 100,
|
152 |
AutoEvalColumn.CII_Bench.name: 100,
|
153 |
AutoEvalColumn.Blink.name: 100,
|
154 |
+
AutoEvalColumn.CharXiv.name: 100,
|
155 |
+
AutoEvalColumn.MathVerse.name: 100,
|
156 |
+
AutoEvalColumn.MmvetV2.name: 100,
|
157 |
+
AutoEvalColumn.Ocrlite.name: 100,
|
158 |
+
AutoEvalColumn.OcrliteZh.name: 100,
|
159 |
AutoEvalColumn.dummy.name: "baseline",
|
160 |
AutoEvalColumn.model_type.name: "",
|
161 |
AutoEvalColumn.flagged.name: False,
|
|
|
172 |
AutoEvalColumn.MMMU.name: 100,
|
173 |
AutoEvalColumn.MMMU_Pro_standard.name: 100,
|
174 |
AutoEvalColumn.MMMU_Pro_vision.name: 100,
|
|
|
175 |
AutoEvalColumn.MathVision.name: 100,
|
176 |
AutoEvalColumn.CII_Bench.name: 100,
|
177 |
AutoEvalColumn.Blink.name: 100,
|
178 |
+
AutoEvalColumn.CharXiv.name: 100,
|
179 |
+
AutoEvalColumn.MathVerse.name: 100,
|
180 |
+
AutoEvalColumn.MmvetV2.name: 100,
|
181 |
+
AutoEvalColumn.Ocrlite.name: 100,
|
182 |
+
AutoEvalColumn.OcrliteZh.name: 100,
|
183 |
AutoEvalColumn.dummy.name: "human_baseline",
|
184 |
AutoEvalColumn.model_type.name: "",
|
185 |
AutoEvalColumn.flagged.name: False,
|