linjunyao
commited on
Commit
•
6734e22
1
Parent(s):
f1fda01
updated citation and intro
Browse files- app.py +2 -2
- judgerbench/meta_data.py +13 -6
app.py
CHANGED
@@ -41,7 +41,7 @@ with gr.Blocks() as demo:
|
|
41 |
# timestamp = struct['time']
|
42 |
|
43 |
# EVAL_TIME = format_timestamp(timestamp)
|
44 |
-
EVAL_TIME = '
|
45 |
|
46 |
# results = struct['results']
|
47 |
# N_MODEL = len(results)
|
@@ -284,7 +284,7 @@ with gr.Blocks() as demo:
|
|
284 |
value=CITATION_BUTTON_TEXT,
|
285 |
label=CITATION_BUTTON_LABEL,
|
286 |
elem_id='citation-button',
|
287 |
-
lines=
|
288 |
)
|
289 |
|
290 |
|
|
|
41 |
# timestamp = struct['time']
|
42 |
|
43 |
# EVAL_TIME = format_timestamp(timestamp)
|
44 |
+
EVAL_TIME = '20241022'
|
45 |
|
46 |
# results = struct['results']
|
47 |
# N_MODEL = len(results)
|
|
|
284 |
value=CITATION_BUTTON_TEXT,
|
285 |
label=CITATION_BUTTON_LABEL,
|
286 |
elem_id='citation-button',
|
287 |
+
lines=7,
|
288 |
)
|
289 |
|
290 |
|
judgerbench/meta_data.py
CHANGED
@@ -21,12 +21,13 @@ STYLE_CLASS_MAPPING = {
|
|
21 |
URL = "http://opencompass.openxlab.space/assets/OpenVLM.json"
|
22 |
VLMEVALKIT_README = 'https://raw.githubusercontent.com/open-compass/VLMEvalKit/main/README.md'
|
23 |
# CONSTANTS-CITATION
|
24 |
-
CITATION_BUTTON_TEXT = r"""@
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
}
|
|
|
30 |
CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
|
31 |
|
32 |
# CONSTANTS-TEXT
|
@@ -34,6 +35,12 @@ LEADERBORAD_INTRODUCTION = """# CompassJudger Subjective Evaluation Learderboard
|
|
34 |
|
35 |
### Welcome to the CompassJudger Subjective Evaluation Learderboard!
|
36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
This leaderboard was last updated: {}.
|
38 |
|
39 |
*All results from the corresponding datasets have been normalized to percentages.
|
|
|
21 |
URL = "http://opencompass.openxlab.space/assets/OpenVLM.json"
|
22 |
VLMEVALKIT_README = 'https://raw.githubusercontent.com/open-compass/VLMEvalKit/main/README.md'
|
23 |
# CONSTANTS-CITATION
|
24 |
+
CITATION_BUTTON_TEXT = r"""@article{cao2024compass,
|
25 |
+
title={CompassJudger-1: All-in-one Judge Model Helps Model Evaluation and Evolution},
|
26 |
+
author={Maosong Cao, Alexander Lam, Haodong Duan, Hongwei Liu, Songyang Zhang, Kai Chen},
|
27 |
+
journal={arXiv preprint arXiv:2410.16256},
|
28 |
+
year={2024}
|
29 |
+
}
|
30 |
+
"""
|
31 |
CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
|
32 |
|
33 |
# CONSTANTS-TEXT
|
|
|
35 |
|
36 |
### Welcome to the CompassJudger Subjective Evaluation Learderboard!
|
37 |
|
38 |
+
Model Link: https://huggingface.co/opencompass/CompassJudger-1-7B-Instruct
|
39 |
+
|
40 |
+
Code Link: https://github.com/open-compass/CompassJudger
|
41 |
+
|
42 |
+
Paper Link: https://arxiv.org/pdf/2410.16256
|
43 |
+
|
44 |
This leaderboard was last updated: {}.
|
45 |
|
46 |
*All results from the corresponding datasets have been normalized to percentages.
|