Reproduction question for SDXL Base 1.0 when reproduce the v2.1 benchmark

[choucaicai]

when I reproduce the v2.1 benchmark，I tested and found SDXL Base 0.9 to be 28.71 and SDXL Base 1.0 to be 28.83, but the scores in the README exceed 30. I used the default settings of diffusers.

[choucaicai]

Even when I ran the given code, I only got an average score of 28.71.

import hpsv2

# print(hpsv2.get_available_models()) # Get models that have access to data
hpsv2.evaluate_benchmark('SDXL Base 0.9')

result：

SDXL-base-0.9 concept-art     29.42      0.2775
SDXL-base-0.9 anime           31.03      0.2606
SDXL-base-0.9 paintings       28.83      0.5296
SDXL-base-0.9 photo           25.56      0.5209
SDXL-base-0.9 Average         28.71 

[Owen718]

+1 Same Question.

[heekhero]

I got the same question, the following are results from my reproduction:
SDXL-base-0.9 concept-art 29.42
SDXL-base-0.9 anime 31.03
SDXL-base-0.9 paintings 28.83
SDXL-base-0.9 photo 25.56
SDXL-base-0.9 Average 28.71
shifting from the reported results.

[heekhero]

Even when I ran the given code, I only got an average score of 28.71.

import hpsv2

# print(hpsv2.get_available_models()) # Get models that have access to data
hpsv2.evaluate_benchmark('SDXL Base 0.9')

result：

SDXL-base-0.9 concept-art     29.42      0.2775
SDXL-base-0.9 anime           31.03      0.2606
SDXL-base-0.9 paintings       28.83      0.5296
SDXL-base-0.9 photo           25.56      0.5209
SDXL-base-0.9 Average         28.71 

same question