Code

Feature/cpu rope#256

Code

Merged

FL33TW00D merged 34 commits intomasterhuggingface/ratchet:masterfrom feature/cpu-ropehuggingface/ratchet:feature/cpu-ropeCopy head branch name to clipboardOct 29, 2024

+376-70

Commits

Commits on Sep 6, 2024

• initial rope setup

  ivarflakstad

  committed
  ab223ca

  Copy full SHA for ab223ca

• tidy

  ivarflakstad

  committed
  2534dfb

  Copy full SHA for 2534dfb

• chore: refactor cpu gemm for use in rope

  ivarflakstad

  committed
  f7d3a32

  Copy full SHA for f7d3a32

• debugging cpu RoPE

  ivarflakstad

  committed
  5f15257

  Copy full SHA for 5f15257

• Rope cpu is almost there

  ivarflakstad

  committed
  70dc153

  Copy full SHA for 70dc153

Commits on Sep 9, 2024

• debugging gemm/rope interaction

  ivarflakstad

  committed
  d1f31da

  Copy full SHA for d1f31da

• More debugging gemm/rope

  ivarflakstad

  committed
  2366e9d

  Copy full SHA for 2366e9d

• Revert gemm back tobinary mul

  ivarflakstad

  committed
  9b4bd6c

  Copy full SHA for 9b4bd6c

• chore: turns out while I was confused, it was not about gemm

  ivarflakstad

  committed
  f998176

  Copy full SHA for f998176

• chore: Add interleave_by_offset

  ivarflakstad

  committed
  a09f986

  Copy full SHA for a09f986

• close

  ivarflakstad

  committed
  46861fa

  Copy full SHA for 46861fa

• Most RoPE test cases are passing

  ivarflakstad

  committed
  23f2687

  Copy full SHA for 23f2687

Commits on Sep 18, 2024

• getting there

  ivarflakstad

  committed
  328d8ce

  Copy full SHA for 328d8ce

Commits on Sep 20, 2024

• Merge branch 'master' into feature/cpu-rope

  ivarflakstad

  committed
  6e39c34

  Copy full SHA for 6e39c34

Commits on Oct 1, 2024

• testing a bunch of different things. really messy :)

  ivarflakstad

  committed
  33b097e

  Copy full SHA for 33b097e

Commits on Oct 2, 2024

• chore: focus on theta

  FL33TW00D

  committed
  d5fb9f8

  Copy full SHA for d5fb9f8

• chore: theta matches

  FL33TW00D

  committed
  44bc1ec

  Copy full SHA for 44bc1ec

• chore: theta matches

  FL33TW00D

  committed
  81f4bfc

  Copy full SHA for 81f4bfc

Commits on Oct 3, 2024

• chore: R1 and R2 match

  FL33TW00D

  committed
  82435eb

  Copy full SHA for 82435eb

• chore: cleaning

  FL33TW00D

  committed
  ca5f5a7

  Copy full SHA for ca5f5a7

Commits on Oct 4, 2024

• chore: RoPE works but is shit

  FL33TW00D

  committed
  88e7c07

  Copy full SHA for 88e7c07

• chore: RoPE doesn't work

  FL33TW00D

  committed
  4d63692

  Copy full SHA for 4d63692

• chore: not quite right

  FL33TW00D

  committed
  1d93205

  Copy full SHA for 1d93205

Commits on Oct 16, 2024

• chore: rope concat dynamic outs length

  ivarflakstad

  committed
  572e7d1

  Copy full SHA for 572e7d1

• chore: simplify rope concat

  ivarflakstad

  committed
  ce991ba

  Copy full SHA for ce991ba

Commits on Oct 18, 2024

• chore: padding r1/r2 with 0s works. Not optimal

  ivarflakstad

  committed
  67a40c9

  Copy full SHA for 67a40c9

Commits on Oct 20, 2024

• chore: use randn in rope test to avoid precision issues

  ivarflakstad

  committed
  c932fd5

  Copy full SHA for c932fd5

• chore: remove redundant "outs" vec

  ivarflakstad

  committed
  022db5a

  Copy full SHA for 022db5a

• chore: use iter cycle instead of % check

  ivarflakstad

  committed
  508b5ed

  Copy full SHA for 508b5ed

Commits on Oct 28, 2024

• Merge branch 'master' into feature/cpu-rope

  ivarflakstad

  committed
  52863d2

  Copy full SHA for 52863d2

• chore: remove unused strided iterator. may be useful later

  ivarflakstad

  committed
  be77442

  Copy full SHA for be77442

Commits on Oct 29, 2024

• chore: tidy up

  ivarflakstad

  committed
  5a018ce

  Copy full SHA for 5a018ce

• chore: ? > unwrap

  ivarflakstad

  committed
  99a074e

  Copy full SHA for 99a074e

• chore: Add back default debug_struct in Debug for Tensor impl

  ivarflakstad

  committed
  b74e4b2

  Copy full SHA for b74e4b2