Code

fused_concat: Support fp8-fp32#122

Code

Open

fischeti wants to merge 4 commits intovivianep/tcasai-benchmarkingpulp-platform/snitch_cluster:vivianep/tcasai-benchmarkingfrom fischeti/tcasaipulp-platform/snitch_cluster:fischeti/tcasaiCopy head branch name to clipboard

+42-26

Commits

Commits on Mar 20, 2024

• fused_concat: Verify with flexfloat/numpy instead of torch

  fischeti

  committed
  7571f7d

• fused_concat: Tranpose the weight matrix for optimized SIMD kernels

  fischeti

  committed
  3aa9d04

• fused_concat: Set more reasonable default parameters

  fischeti

  committed
  f0f2ce2

Commits on Mar 21, 2024

• fused_concat: Reduce error threshold for fp16

  fischeti

  committed
  68c718a