From 1169ffa4800304f894af662b911318af869302aa Mon Sep 17 00:00:00 2001 From: Junkun Date: Thu, 17 Feb 2022 22:03:47 -0800 Subject: [PATCH] add config files --- examples/mustc/st1/conf/transformer_de.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_es.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_fr.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_it.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_nl.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_pt.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_ro.yaml | 90 +++++++++++++++++++++ examples/mustc/st1/conf/transformer_ru.yaml | 90 +++++++++++++++++++++ 8 files changed, 720 insertions(+) create mode 100644 examples/mustc/st1/conf/transformer_de.yaml create mode 100644 examples/mustc/st1/conf/transformer_es.yaml create mode 100644 examples/mustc/st1/conf/transformer_fr.yaml create mode 100644 examples/mustc/st1/conf/transformer_it.yaml create mode 100644 examples/mustc/st1/conf/transformer_nl.yaml create mode 100644 examples/mustc/st1/conf/transformer_pt.yaml create mode 100644 examples/mustc/st1/conf/transformer_ro.yaml create mode 100644 examples/mustc/st1/conf/transformer_ru.yaml diff --git a/examples/mustc/st1/conf/transformer_de.yaml b/examples/mustc/st1/conf/transformer_de.yaml new file mode 100644 index 00000000..4e498853 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_de.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.de.train +dev_manifest: data/manifest.de.dev +test_manifest: data/manifest.de.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-de.de_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-de.de_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_es.yaml b/examples/mustc/st1/conf/transformer_es.yaml new file mode 100644 index 00000000..4f4e1e52 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_es.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.es.train +dev_manifest: data/manifest.es.dev +test_manifest: data/manifest.es.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-es.es_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-es.es_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_fr.yaml b/examples/mustc/st1/conf/transformer_fr.yaml new file mode 100644 index 00000000..ede10364 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_fr.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.fr.train +dev_manifest: data/manifest.fr.dev +test_manifest: data/manifest.fr.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-fr.fr_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-fr.fr_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_it.yaml b/examples/mustc/st1/conf/transformer_it.yaml new file mode 100644 index 00000000..265aebb2 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_it.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.it.train +dev_manifest: data/manifest.it.dev +test_manifest: data/manifest.it.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-it.it_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-it.it_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_nl.yaml b/examples/mustc/st1/conf/transformer_nl.yaml new file mode 100644 index 00000000..47f44500 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_nl.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.nl.train +dev_manifest: data/manifest.nl.dev +test_manifest: data/manifest.nl.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-nl.nl_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-nl.nl_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_pt.yaml b/examples/mustc/st1/conf/transformer_pt.yaml new file mode 100644 index 00000000..40f77843 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_pt.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.pt.train +dev_manifest: data/manifest.pt.dev +test_manifest: data/manifest.pt.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-pt.pt_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-pt.pt_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_ro.yaml b/examples/mustc/st1/conf/transformer_ro.yaml new file mode 100644 index 00000000..eff2960f --- /dev/null +++ b/examples/mustc/st1/conf/transformer_ro.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.ro.train +dev_manifest: data/manifest.ro.dev +test_manifest: data/manifest.ro.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-ro.ro_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-ro.ro_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file diff --git a/examples/mustc/st1/conf/transformer_ru.yaml b/examples/mustc/st1/conf/transformer_ru.yaml new file mode 100644 index 00000000..83f98f57 --- /dev/null +++ b/examples/mustc/st1/conf/transformer_ru.yaml @@ -0,0 +1,90 @@ +# https://yaml.org/type/float.html +########################################### +# Data # +########################################### +train_manifest: data/manifest.ru.train +dev_manifest: data/manifest.ru.dev +test_manifest: data/manifest.ru.test + +########################################### +# Dataloader # +########################################### +vocab_filepath: data/lang_1spm/train_sp.en-ru.ru_bpe8000_units_tc.txt +unit_type: 'spm' +spm_model_prefix: data/lang_1spm/train_sp.en-ru.ru_bpe8000_tc +mean_std_filepath: "" +# preprocess_config: conf/augmentation.json +batch_size: 20 +feat_dim: 83 +stride_ms: 10.0 +window_ms: 25.0 +sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs +maxlen_in: 512 # if input length > maxlen-in, batchsize is automatically reduced +maxlen_out: 150 # if output length > maxlen-out, batchsize is automatically reduced +minibatches: 0 # for debug +batch_count: auto +batch_bins: 0 +batch_frames_in: 0 +batch_frames_out: 0 +batch_frames_inout: 0 +preprocess_config: +num_workers: 0 +subsampling_factor: 1 +num_encs: 1 + + +############################################ +# Network Architecture # +############################################ +cmvn_file: None +cmvn_file_type: "json" +# encoder related +encoder: transformer +encoder_conf: + output_size: 256 # dimension of attention + attention_heads: 4 + linear_units: 2048 # the number of units of position-wise feed forward + num_blocks: 12 # the number of encoder blocks + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + attention_dropout_rate: 0.0 + input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8 + normalize_before: true + +# decoder related +decoder: transformer +decoder_conf: + attention_heads: 4 + linear_units: 2048 + num_blocks: 6 + dropout_rate: 0.1 + positional_dropout_rate: 0.1 + self_attention_dropout_rate: 0.0 + src_attention_dropout_rate: 0.0 + +# hybrid CTC/attention +model_conf: + asr_weight: 0.0 + ctc_weight: 0.0 + lsm_weight: 0.1 # label smoothing option + length_normalized_loss: false + + +########################################### +# Training # +########################################### +n_epoch: 40 +accum_grad: 2 +global_grad_clip: 5.0 +optim: adam +optim_conf: + lr: 2.5 + weight_decay: 0. +scheduler: noam +scheduler_conf: + warmup_steps: 25000 + lr_decay: 1.0 +log_interval: 50 +checkpoint: + kbest_n: 50 + latest_n: 5 \ No newline at end of file -- GitLab