Code

Future dev#2

Code

Open

jkiesele wants to merge 31 commits intokdlong:multigpukdlong/HGCalML:multigpufrom jkiesele:future_devjkiesele/HGCalML-1:future_devCopy head branch name to clipboard

+1,818-531

Commits

Commits on Dec 18, 2023

• training works now with ragged inputs. missing: distribute strategy and callback - the latter not that important

  jkiesele

  committed
  ac1bd0c

  Copy full SHA for ac1bd0c

• clean up

  jkiesele

  committed
  a92c7d0

  Copy full SHA for a92c7d0

Commits on Dec 19, 2023

• snapshot

  jkiesele

  committed
  e9a39e3

  Copy full SHA for e9a39e3

• revert for fresh start

  jkiesele

  committed
  0b8fae9

  Copy full SHA for 0b8fae9

• working mgpu training (no callbacks)

  jkiesele

  committed
  4464bd8

  Copy full SHA for 4464bd8

Commits on Dec 20, 2023

• added compatibility

  jkiesele

  committed
  c602aff

  Copy full SHA for c602aff

• training callbacks are back

  jkiesele

  committed
  e200fe3

  Copy full SHA for e200fe3

Commits on Dec 29, 2023

• status update for tf2.14

  jkiesele

  committed
  788dad0

  Copy full SHA for 788dad0

• compat between tf2.4 and tf2.14

  jkiesele

  committed
  7c805b1

  Copy full SHA for 7c805b1

• compile error

  jkiesele

  committed
  e841ce3

  Copy full SHA for e841ce3

Commits on Dec 31, 2023

• fixes for tf2.14 and new cuda (may have affected earlier trainings already, careful)

  jkiesele

  committed
  46e379f

  Copy full SHA for 46e379f

• debug switch off

  jkiesele

  committed
  0e4c084

  Copy full SHA for 0e4c084

Commits on Jan 1, 2024

• getting ready for new training

  jkiesele

  committed
  f412471

  Copy full SHA for f412471

• dead zone hinge and pushed batch size (for 80gb A100)

  jkiesele

  committed
  57dc99b

  Copy full SHA for 57dc99b

Commits on Jan 2, 2024

• moved to new djc changes and put wandb in the relevant places

  jkiesele

  committed
  df6bf73

  Copy full SHA for df6bf73

• readme

  jkiesele

  committed
  c13a486

  Copy full SHA for c13a486

• some perf tweaks ans small fixes

  jkiesele

  committed
  0218110

  Copy full SHA for 0218110

• some performance increasing bits

  jkiesele

  committed
  ee520b3

  Copy full SHA for ee520b3

Commits on Jan 8, 2024

• some more fixes

  jkiesele

  committed
  d7e9060

  Copy full SHA for d7e9060

• compat

  jkiesele

  committed
  fec58f2

  Copy full SHA for fec58f2

• wandb metrics

  jkiesele

  committed
  6729aa6

  Copy full SHA for 6729aa6

Commits on Jan 13, 2024

• convenience function - untested

  jkiesele

  committed
  8ee7015

  Copy full SHA for 8ee7015

• small helper now tested

  jkiesele

  committed
  e123e44

  Copy full SHA for e123e44

• more docu

  jkiesele

  committed
  db76678

  Copy full SHA for db76678

Commits on Jan 22, 2024

• small fixes

  jkiesele

  committed
  d3387b5

  Copy full SHA for d3387b5

• fraction regressor loss

  jkiesele

  committed
  a28850c

  Copy full SHA for a28850c

Commits on Jan 23, 2024

• small fix

  jkiesele

  committed
  35905f2

  Copy full SHA for 35905f2

• new reduce layer

  jkiesele

  committed
  ee3202d

  Copy full SHA for ee3202d

• new reduce layer fix

  jkiesele

  committed
  e1917b3

  Copy full SHA for e1917b3

• another layer

  jkiesele

  committed
  2dc561a

  Copy full SHA for 2dc561a

Commits on Feb 12, 2024

• Future dev (#8)

  jkiesele

  authored
  63fdeec

  Copy full SHA for 63fdeec