Code

Various fixes#332

Code

Merged

percevalw merged 12 commits intomasteraphp/edsnlp:masterfrom various-fixesaphp/edsnlp:various-fixesCopy head branch name to clipboardNov 14, 2024

+875-417

Commits

Commits on Nov 13, 2024

• docs: improve data docs

  percevalw

  committed
  ffcfc4a

• feat: improve optim scheduling by setting schedules in place of values

  percevalw

  committed
  38d1ad5

• fix: improve pydantic v1/v2 support

  percevalw

  committed
  4392bd4

• fix: bound both input and output queues in mp to avoid pipeline congestion

  percevalw

  committed
  32c1cd1

• fix: support specifying only 'components' when defining a pipeline

  percevalw

  committed
  5ffb556

• fix: remove unused, undocumented and useless options & warnings

  percevalw

  committed
  9c73d17

• refacto: rename @pipelines->@core, optim->optimizer, explicitly instanciate pipeline in train configs

  percevalw

  committed
  eaeb1d1

• perf: prune tensors in preprocessing workers during training

  percevalw

  committed
  77e5e8e

• feat: support custom forward use in torch components + generic batch_to_device

  percevalw

  committed
  fe63753

• fix: re-enable training multi-task model with shared weights

  percevalw

  committed
  83a2538

• refacto: rename @pipelines->@core, optim->optimizer, explicitly instanciate pipeline in train configs

  percevalw

  committed
  bcd7359

• fix: support multi-gpu training

  percevalw

  committed
  a8654f5