Discrepancy Between DDIM CIFAR-10 Results and Paper

[IC-Yuan]

Thank you for providing the code! I used the cifar_w4a8_ckpt.pth you provided to directly generate images using the following command:

python scripts/sample_diffusion_ddim.py --config configs/cifar10.yml --use_pretrained --timesteps 100 --eta 0 --skip_type quad --ptq --weight_bit 4 --quant_mode qdiff --quant_act --act_bit 8 --a_sym --split --resume -l <output_path> --cali_ckpt <quantized_ckpt_path>

The resulting FID score is 4.79, which differs from the 4.93 reported in the paper for "DDIM on CIFAR-10 W4A8." I am wondering if this discrepancy might be due to an issue with my dataset. I am using the CIFAR-10 training set, which contains a total of 50,000 images. Could you help clarify?