feat: lr scheduler by Chamberlain0w0 · Pull Request #151 · InfiniTensor/InfiniTrain

refactor(optimizer): hoist learning_rate_ to Optimizer base and add lr accessors, passthrough SetLearningRate/GetLearningRate, and add initial_learning_rate and it's accessors

kinorw

authored and

Chamberlain0w0

committed

feat(lr_scheduler): add LRScheduler abstract base class, ConstantLR, StepLR, LinearLR, LambdaLR and SequentialLR

kinorw

authored and

Chamberlain0w0

committed

refactor(lr_scheduler): apply template method pattern to LRScheduler base class, add factory method Create<T>() with two-phase init and update all tests to use Create<T>() factory method.

kinorw

authored and

Chamberlain0w0

committed

feat(lr_scheduler): add ChainedScheduler composite strategy, and add closed and chained form, adjust LinearLR、SequentialLR

kinorw

authored and

Chamberlain0w0

committed

feat(lr_scheduler): add scheduler factory for CLI integration

kinorw

authored and

Chamberlain0w0

committed

refactor(examples): add scheduler placeholder and use runtime lr in logs, and integrate scheduler into training loop

kinorw

authored and

Chamberlain0w0

committed

style: apply clang-format to all legacy files

littleotherut

authored and

Chamberlain0w0

committed

refactor: rename current_lr_ to recover_lr_ and update related methods, add validation tests for learning rate schedulers

kinorw

authored and

Chamberlain0w0

committed

fix: adapt to megatron-style arguments

Chamberlain0w0

committed

style: fix setter/getter name

Chamberlain0w0

committed

fix: add lr_scheduler test group in config

Chamberlain0w0

committed

fix: resolve comments

Chamberlain0w0

committed

refactor: reorganize unit tests to comply with ctest infra

Chamberlain0w0

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: lr scheduler#151

feat: lr scheduler#151
Chamberlain0w0 wants to merge 13 commits into
masterfrom
feat/lr_scheduler

Commits on May 21, 2026

refactor(optimizer): hoist learning_rate_ to Optimizer base and add lr accessors, passthrough SetLearningRate/GetLearningRate, and add initial_learning_rate and it's accessors

feat(lr_scheduler): add LRScheduler abstract base class, ConstantLR, StepLR, LinearLR, LambdaLR and SequentialLR

refactor(lr_scheduler): apply template method pattern to LRScheduler base class, add factory method Create<T>() with two-phase init and update all tests to use Create<T>() factory method.

feat(lr_scheduler): add ChainedScheduler composite strategy, and add closed and chained form, adjust LinearLR、SequentialLR

feat(lr_scheduler): add scheduler factory for CLI integration

refactor(examples): add scheduler placeholder and use runtime lr in logs, and integrate scheduler into training loop

style: apply clang-format to all legacy files

refactor: rename current_lr_ to recover_lr_ and update related methods, add validation tests for learning rate schedulers

fix: adapt to megatron-style arguments

style: fix setter/getter name

fix: add lr_scheduler test group in config

fix: resolve comments

refactor: reorganize unit tests to comply with ctest infra