Code

Feature/cpu gemm#247

Code

Merged

FL33TW00D merged 18 commits intomasterhuggingface/ratchet:masterfrom feature/cpu-gemmhuggingface/ratchet:feature/cpu-gemmCopy head branch name to clipboardAug 24, 2024

+581-253

Commits

Commits on Aug 15, 2024

• chore: initial cpu gemm impl

  ivarflakstad

  committed
  a17add9

  Copy full SHA for a17add9

• chore: no idea

  ivarflakstad

  committed
  b5457d0

  Copy full SHA for b5457d0

• chore: gemm working except TransposeKind::Out

  ivarflakstad

  committed
  b75dd80

  Copy full SHA for b75dd80

Commits on Aug 16, 2024

• chore: out -> dst

  ivarflakstad

  committed
  f59575a

  Copy full SHA for f59575a

• chore: tidy m, n, and k helpers. Begin trans_dst work

  ivarflakstad

  committed
  f848035

  Copy full SHA for f848035

• chore: tidy

  ivarflakstad

  committed
  97d8abe

  Copy full SHA for 97d8abe

• chore: use iter (enumerate) take(n) instead of direct indexing

  ivarflakstad

  committed
  f92365e

  Copy full SHA for f92365e

• fix: avoid redundant transposes. input and dst transposes + (a b)T => bT aT rule leads to XOR operator

  ivarflakstad

  committed
  9c6f954

  Copy full SHA for 9c6f954

• temp stash

  ivarflakstad

  committed
  4c83595

  Copy full SHA for 4c83595

• fix: gemm works! <insert goat emoji here>

  ivarflakstad

  committed
  51f1c0b

  Copy full SHA for 51f1c0b

• chore: separate test_sgemm_gpu/cpu tests

  ivarflakstad

  committed
  fea7140

  Copy full SHA for fea7140

• chore: unify gemm naming conventions

  ivarflakstad

  committed
  6537acf

  Copy full SHA for 6537acf

Commits on Aug 17, 2024

• chore: debugging gpu matmul kernels

  ivarflakstad

  committed
  f4e4173

  Copy full SHA for f4e4173

Commits on Aug 23, 2024

• fix: gemm works on both cpu and gpu

  ivarflakstad

  committed
  35eca06

  Copy full SHA for 35eca06

• chore: 🧹

  ivarflakstad

  committed
  cba87ed

  Copy full SHA for cba87ed

• chore: remove redundant test

  ivarflakstad

  committed
  eb206a6

  Copy full SHA for eb206a6

• chore: disable test of cpu index_select for quantized tensors

  ivarflakstad

  committed
  4cda652

  Copy full SHA for 4cda652

Commits on Aug 24, 2024

• feat: f16 and bf16 gemm support

  ivarflakstad

  committed
  e9bc238

  Copy full SHA for e9bc238