Code

Allow any model architecture#3

Code

Open

ClashLuke wants to merge 22 commits intomaindvruette/locoprop-torch:mainfrom dual-forwarddvruette/locoprop-torch:dual-forwardCopy head branch name to clipboard

+446-125

Commits

Commits on Jan 15, 2023

• feat(locoprop): use lightning for trainer

  ClashLuke

  committed
  86b489a

  Copy full SHA for 86b489a

• feat(locoprop): use lightning for trainer v2

  ClashLuke

  committed
  672227a

  Copy full SHA for 672227a

• style(locoprop): set optimizer params in LocopropTrainer

  ClashLuke

  committed
  c3a3778

  Copy full SHA for c3a3778

• style(locoprop): remove PL dependency

  ClashLuke

  committed
  7843413

  Copy full SHA for 7843413

• Merge branch 'main' into dual-forward

  Show description for 6119b41
  ClashLuke

  committed
  6119b41

  Copy full SHA for 6119b41

• merge main

  ClashLuke

  committed
  7a30fac

  Copy full SHA for 7a30fac

Commits on Jan 16, 2023

• style(layer): comment _IS_TRAINING

  Show description for b859a48
  

  ClashLuke

  and

  dvruette

  authored
  b859a48

  Copy full SHA for b859a48

• fix(layer): only take std within sample

  ClashLuke

  authored
  b0e0d2a

  Copy full SHA for b0e0d2a

• fix correction magnitude

  dvruette

  committed
  e28e425

  Copy full SHA for e28e425

• feat(layer): remove eps

  ClashLuke

  authored
  1ab679b

  Copy full SHA for 1ab679b

• fix(trainer): dont set removed eps

  ClashLuke

  authored
  76ddf30

  Copy full SHA for 76ddf30

Commits on Mar 25, 2024

• test(locolayer): use lerp

  ClashLuke

  committed
  255fff4

  Copy full SHA for 255fff4

• test(model): add LocoLinear

  ClashLuke

  committed
  d3179b4

  Copy full SHA for d3179b4

• fix(trainer): correct IS_BASELINE usage

  ClashLuke

  committed
  4c10742

  Copy full SHA for 4c10742

• style: add example notebook

  ClashLuke

  committed
  4ab73a0

  Copy full SHA for 4ab73a0

• fix(layer): batchsize invariant

  ClashLuke

  committed
  8bcea2f

  Copy full SHA for 8bcea2f

• perf(locoprop): use different optimizer, add torch flags

  ClashLuke

  committed
  281da8a

  Copy full SHA for 281da8a

Commits on Mar 26, 2024

• perf(layer): use zero memory tensor

  ClashLuke

  committed
  1efe641

  Copy full SHA for 1efe641

• style(example): more notebook to python script

  ClashLuke

  committed
  186220b

  Copy full SHA for 186220b

• fix(layer): don't use torch fn in unwrap

  ClashLuke

  committed
  71b7056

  Copy full SHA for 71b7056

• fix(layer): allow normal loco

  ClashLuke

  committed
  f600a22

  Copy full SHA for f600a22

• fix(layer): now

  ClashLuke

  committed
  2a6e8e1

  Copy full SHA for 2a6e8e1