diff options
Diffstat (limited to 'training')
| -rw-r--r-- | training/conf/config.yaml | 4 | ||||
| -rw-r--r-- | training/conf/experiment/conv_transformer_lines.yaml | 8 | ||||
| -rw-r--r-- | training/conf/experiment/conv_transformer_paragraphs.yaml | 8 | ||||
| -rw-r--r-- | training/conf/lr_scheduler/cosine_annealing.yaml (renamed from training/conf/lr_schedulers/cosine_annealing.yaml) | 0 | ||||
| -rw-r--r-- | training/conf/lr_scheduler/one_cycle.yaml (renamed from training/conf/lr_schedulers/one_cycle.yaml) | 0 | ||||
| -rw-r--r-- | training/conf/lr_scheduler/reduce_on_plateau.yaml (renamed from training/conf/lr_schedulers/reduce_on_plateau.yaml) | 0 | ||||
| -rw-r--r-- | training/conf/optimizer/radam.yaml (renamed from training/conf/optimizers/radam.yaml) | 0 | ||||
| -rw-r--r-- | training/run.py | 4 |
8 files changed, 12 insertions, 12 deletions
diff --git a/training/conf/config.yaml b/training/conf/config.yaml index fdeb5d2..46a1e43 100644 --- a/training/conf/config.yaml +++ b/training/conf/config.yaml @@ -7,10 +7,10 @@ defaults: - datamodule: iam_extended_paragraphs - hydra: default - logger: wandb - - lr_schedulers: cosine_annealing + - lr_scheduler: cosine_annealing - model: lit_transformer - network: conv_transformer - - optimizers: radam + - optimizer: radam - trainer: default seed: 4711 diff --git a/training/conf/experiment/conv_transformer_lines.yaml b/training/conf/experiment/conv_transformer_lines.yaml index 48df78d..eb9bc9e 100644 --- a/training/conf/experiment/conv_transformer_lines.yaml +++ b/training/conf/experiment/conv_transformer_lines.yaml @@ -6,8 +6,8 @@ defaults: - override /datamodule: iam_lines - override /network: conv_transformer - override /model: lit_transformer - - override /lr_schedulers: null - - override /optimizers: null + - override /lr_scheduler: null + - override /optimizer: null epochs: &epochs 512 ignore_index: &ignore_index 3 @@ -28,7 +28,7 @@ callbacks: annealing_strategy: cos device: null -optimizers: +optimizer: _target_: torch.optim.RAdam lr: 3.0e-4 betas: [0.9, 0.999] @@ -36,7 +36,7 @@ optimizers: eps: 1.0e-8 parameters: network -lr_schedulers: +lr_scheduler: _target_: torch.optim.lr_scheduler.OneCycleLR max_lr: 3.0e-4 total_steps: null diff --git a/training/conf/experiment/conv_transformer_paragraphs.yaml b/training/conf/experiment/conv_transformer_paragraphs.yaml index d0d0314..7c6e231 100644 --- a/training/conf/experiment/conv_transformer_paragraphs.yaml +++ b/training/conf/experiment/conv_transformer_paragraphs.yaml @@ -6,8 +6,8 @@ defaults: - override /datamodule: iam_extended_paragraphs - override /network: conv_transformer - override /model: lit_transformer - - override /lr_schedulers: null - - override /optimizers: null + - override /lr_scheduler: null + - override /optimizer: null epochs: &epochs 600 num_classes: &num_classes 58 @@ -28,7 +28,7 @@ callbacks: annealing_strategy: cos device: null -optimizers: +optimizer: _target_: torch.optim.RAdam lr: 3.0e-4 betas: [0.9, 0.999] @@ -36,7 +36,7 @@ optimizers: eps: 1.0e-8 parameters: network -lr_schedulers: +lr_scheduler: _target_: torch.optim.lr_scheduler.OneCycleLR max_lr: 3.0e-4 total_steps: null diff --git a/training/conf/lr_schedulers/cosine_annealing.yaml b/training/conf/lr_scheduler/cosine_annealing.yaml index e8364f0..e8364f0 100644 --- a/training/conf/lr_schedulers/cosine_annealing.yaml +++ b/training/conf/lr_scheduler/cosine_annealing.yaml diff --git a/training/conf/lr_schedulers/one_cycle.yaml b/training/conf/lr_scheduler/one_cycle.yaml index 20eab9f..20eab9f 100644 --- a/training/conf/lr_schedulers/one_cycle.yaml +++ b/training/conf/lr_scheduler/one_cycle.yaml diff --git a/training/conf/lr_schedulers/reduce_on_plateau.yaml b/training/conf/lr_scheduler/reduce_on_plateau.yaml index 9aa9de4..9aa9de4 100644 --- a/training/conf/lr_schedulers/reduce_on_plateau.yaml +++ b/training/conf/lr_scheduler/reduce_on_plateau.yaml diff --git a/training/conf/optimizers/radam.yaml b/training/conf/optimizer/radam.yaml index d11fcb5..d11fcb5 100644 --- a/training/conf/optimizers/radam.yaml +++ b/training/conf/optimizer/radam.yaml diff --git a/training/run.py b/training/run.py index 0e17b4e..68cedc7 100644 --- a/training/run.py +++ b/training/run.py @@ -41,8 +41,8 @@ def run(config: DictConfig) -> Optional[float]: network=network, mapping=datamodule.mapping, loss_fn=loss_fn, - optimizer_configs=config.optimizers, - lr_scheduler_configs=config.lr_schedulers, + optimizer_config=config.optimizer, + lr_scheduler_config=config.lr_scheduler, _recursive_=False, ) |