Deep learning compiler

Block Sparsity on TVM -- 50% spasity = 1.6 speed up + < 1% accuracy loss

Realize efficient sparsity with 50% sparsity by TVM.

Ruihao Gong

Aug 26, 2021 1 min read Deep learning compiler, Sparsity

Block Sparsity on TVM -- 50% spasity = 1.6 speed up + < 1% accuracy loss

Int8 ViT on TVM, 1.5 speed up compared with TensorRT

The first to support Int8 ViT for TVM, achieving a significant speed up.

Ruihao Gong

Apr 19, 2021 1 min read Deep learning compiler, Quantization, Transformer