README里面的【Speedup of Transformer Inference】是哪个模型的，这边用里面的VIT用例测试的性能只有2.64倍？

[Wayne-Bfx]

导出模型：python3 export/huggingface/hf_vit_export.py 测试性能：python3 test/ls_vit.py 环境： V100服务器，torch1.10， python3.7.13，CUDA11.3 性能测试结果：

=========lightseq========= lightseq generating... lightseq time: 0.005731139099225402s lightseq results (class predictions): [1] =========huggingface========= huggingface generating... huggingface time: 0.015130232088267803s huggingface results (class predictions): [1]

[hexisyztem]

"Speedup of Transformer Inference" is not the speedup of vit model

[Wayne-Bfx]

您好，可以给下【Speedup of Transformer Inference】测试的模型名称，或者该模型的github地址不

[hexisyztem]

here: https://github.com/bytedance/lightseq/blob/master/examples/inference/python/export/huggingface/hf_bart_export.py https://github.com/bytedance/lightseq/blob/master/examples/inference/python/test/ls_bart.py

[Wayne-Bfx]

非常感谢