optimize the shell scripts

egs/librispeech/asr/conf/base.yaml

@@ -24,7 +24,7 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-kernel-sizes: 5,5

egs/librispeech/asr/conf/conformer.yaml

@@ -2,4 +2,4 @@ arch: s2t_conformer_s
 
 macaron-style: True
 use-cnn-module: True
-cnn-module-kernel: 15
+cnn-module-kernel: 31

egs/librispeech/asr/conf/ctc.yaml

-train-subset: train-clean-100,train-clean-360,train-other-500
-valid-subset: dev-clean
-
-max-epoch: 100
-max-update: 300000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
-arch: s2t_transformer_s
-share-decoder-input-output-embed: True
-optimizer: adam
-clip-norm: 10.0
-lr-scheduler: inverse_sqrt
-warmup-init-lr: 1e-7
-warmup-updates: 10000
-lr: 2e-3
-#adam_betas: (0.9,0.98)
-
 ctc-weight: 0.3
-criterion: label_smoothed_cross_entropy_with_ctc
-label_smoothing: 0.1
-
-conv-kernel-sizes: 5,5
-conv-channels: 1024
-dropout: 0.1
-activation-fn: relu
-encoder-embed-dim: 256
-encoder-ffn-embed-dim: 2048
-encoder-layers: 12
-decoder-layers: 6
-encoder-attention-heads: 4
-
-decoder-embed-dim: 256
-decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
-attention-dropout: 0.1
-activation-dropout: 0.1

egs/librispeech/asr/conf/ctc_debug.yaml → egs/librispeech/asr/conf/debug.yaml

-#train-subset: train-clean-100,train-clean-360,train-other-500
 train-subset: train-clean-100
 valid-subset: dev-clean
 
 max-epoch: 100
 max-update: 300000
 
-num-workers: 0
+num-workers: 8
 patience: 10
 no-progress-bar: True
 log-interval: 100
 seed: 1
 report-accuracy: True
 
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
+
 arch: s2t_transformer_s
 share-decoder-input-output-embed: True
 optimizer: adam
@@ -22,26 +24,20 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-ctc-weight: 0.3
 criterion: label_smoothed_cross_entropy_with_ctc
+ctc-weight: 0.3
 label_smoothing: 0.1
 
 conv-kernel-sizes: 5,5
 conv-channels: 1024
 dropout: 0.1
 activation-fn: relu
-#encoder-embed-dim: 256
+encoder-embed-dim: 256
 encoder-ffn-embed-dim: 2048
 encoder-layers: 12
-decoder-layers: 3
+decoder-layers: 6
 encoder-attention-heads: 4
 
-macaron-style: True
-use-cnn-module: True
-cnn-module-kernel: 31
-
-adpater: subsample
-
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-attention-heads: 4

egs/mustc/st/conf/pyramid.yaml → egs/librispeech/asr/conf/large.yaml

-arch: pys2t_transformer_s
-#conv-kernel-sizes: 5
-encoder-embed-dim: 512
-pyramid-layers: 3_6_9
-ctc-layer: 7
-
-train-subset: train_st
-valid-subset: dev_st
+train-subset: train-clean-100,train-clean-360,train-other-500
+valid-subset: dev-clean
 
-max-epoch: 50
-max-update: 100000
+max-epoch: 100
+max-update: 300000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -20,6 +14,7 @@ report-accuracy: True
 #load-pretrained-encoder-from:
 #load-pretrained-decoder-from:
 
+arch: s2t_transformer_m
 share-decoder-input-output-embed: True
 optimizer: adam
 clip-norm: 10.0
@@ -29,20 +24,21 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-ctc-weight: 0.3
 criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
+conv-kernel-sizes: 5,5
 conv-channels: 1024
 dropout: 0.1
 activation-fn: relu
+encoder-embed-dim: 512
 encoder-ffn-embed-dim: 2048
 encoder-layers: 12
 decoder-layers: 6
-encoder-attention-heads: 4
+encoder-attention-heads: 8
 
-decoder-embed-dim: 256
+decoder-embed-dim: 512
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 8
 attention-dropout: 0.1
 activation-dropout: 0.1

egs/librispeech/asr/conf/pyramid_stage3.yaml → egs/librispeech/asr/conf/pyramid3_base.yaml

 arch: pys2t_transformer_s
-encoder-embed-dim: 512
+encoder-embed-dim: 256
 pyramid-stages: 3
 pyramid-layers: 3_6_3
-#encoder-attention-type: reduced
-#pyramid-attn-sample-ratios: 4_2_1
-#pyramid-block-attn: True
-#pyramid-fuse-way: gated
-pyramid-use-ppm: True
+pyramid-fuse-way: all_conv
+pyramid-fuse: True
 pyramid-sr-ratios: 2_2_2
-pyramid-embed-dims: 128_256_512
+pyramid-embed-dims: 256_256_256
 pyramid-reduced-embed: conv
 pyramid-embed-norm: True
 pyramid-position-embed: 1_1_1
 pyramid-kernel-sizes: 5_5_5
-pyramid-ffn-ratios: 8_8_4
-pyramid-heads: 2_4_8
+pyramid-ffn-ratios: 8_8_8
+pyramid-heads: 4_4_4
 
 train-subset: train-clean-100,train-clean-360,train-other-500
 valid-subset: dev-clean
@@ -23,7 +20,7 @@ max-epoch: 100
 max-update: 300000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -42,7 +39,6 @@ lr: 2e-3
 #adam_betas: (0.9,0.98)
 
 criterion: label_smoothed_cross_entropy_with_ctc
-ctc-weight: 0
 label_smoothing: 0.1
 
 conv-channels: 1024

egs/mustc/asr/conf/pyramid_stage3.yaml → egs/librispeech/asr/conf/pyramid4_base.yaml

 arch: pys2t_transformer_s
-encoder-embed-dim: 512
-pyramid-stages: 3
-pyramid-layers: 3_6_3
-pyramid-sr-ratios: 2_2_2
-#pyramid-block-attn: True
-#pyramid-fuse-way: add
-pyramid-use-ppm: True
-pyramid-embed-dims: 128_256_512
+encoder-embed-dim: 256
+pyramid-stages: 4
+#pyramid-dropout: 0
+pyramid-layers: 2_2_6_2
+pyramid-sr-ratios: 2_2_2_2
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 256_256_256_256
 pyramid-reduced-embed: conv
 pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1
-pyramid-kernel-sizes: 5_5_5
-pyramid-ffn-ratios: 8_8_4
-pyramid-heads: 2_4_8
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
 
-train-subset: train_asr
-valid-subset: dev_asr
+train-subset: train-clean-100,train-clean-360,train-other-500
+valid-subset: dev-clean
 
-max-epoch: 50
-max-update: 100000
+max-epoch: 100
+max-update: 300000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -39,7 +39,7 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-channels: 1024

egs/librispeech/asr/conf/pyramid4_base_sr8.yaml

+arch: pys2t_transformer_s
+encoder-embed-dim: 256
+pyramid-stages: 4
+pyramid-layers: 3_3_3_3
+pyramid-sr-ratios: 2_2_1_2
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 256_256_256_256
+pyramid-reduced-embed: conv
+pyramid-embed-norm: True
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
+
+train-subset: train-clean-100,train-clean-360,train-other-500
+valid-subset: dev-clean
+
+max-epoch: 100
+max-update: 300000
+
+num-workers: 8
+patience: 20
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
+
+share-decoder-input-output-embed: True
+optimizer: adam
+clip-norm: 10.0
+lr-scheduler: inverse_sqrt
+warmup-init-lr: 1e-7
+warmup-updates: 10000
+lr: 2e-3
+#adam_betas: (0.9,0.98)
+
+criterion: label_smoothed_cross_entropy_with_ctc
+label_smoothing: 0.1
+
+conv-channels: 1024
+dropout: 0.1
+activation-fn: relu
+encoder-ffn-embed-dim: 2048
+encoder-layers: 12
+decoder-layers: 6
+encoder-attention-heads: 4
+
+decoder-embed-dim: 256
+decoder-ffn-embed-dim: 2048
+decoder-attention-heads: 4
+attention-dropout: 0.1
+activation-dropout: 0.1

egs/librispeech/asr/conf/pyramid4_deep18.yaml

+arch: pys2t_transformer_s
+encoder-embed-dim: 256
+pyramid-stages: 4
+pyramid-layers: 3_3_8_4
+pyramid-sr-ratios: 2_2_2_2
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 256_256_256_256
+pyramid-reduced-embed: conv
+pyramid-embed-norm: True
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
+
+train-subset: train-clean-100,train-clean-360,train-other-500
+valid-subset: dev-clean
+
+max-epoch: 100
+max-update: 300000
+
+num-workers: 8
+patience: 20
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
+
+share-decoder-input-output-embed: True
+optimizer: adam
+clip-norm: 10.0
+lr-scheduler: inverse_sqrt
+warmup-init-lr: 1e-7
+warmup-updates: 10000
+lr: 2e-3
+#adam_betas: (0.9,0.98)
+
+criterion: label_smoothed_cross_entropy_with_ctc
+label_smoothing: 0.1
+
+conv-channels: 1024
+dropout: 0.1
+activation-fn: relu
+encoder-ffn-embed-dim: 2048
+encoder-layers: 12
+decoder-layers: 6
+encoder-attention-heads: 4
+
+decoder-embed-dim: 256
+decoder-ffn-embed-dim: 2048
+decoder-attention-heads: 4
+attention-dropout: 0.1
+activation-dropout: 0.1

egs/librispeech/asr/conf/pyramid_enc6.yaml → egs/librispeech/asr/conf/pyramid4_deep30.yaml

 arch: pys2t_transformer_s
-encoder-embed-dim: 512
-pyramid-stages: 3
-pyramid-layers: 3_6_3
-encoder-attention-type: reduced
-pyramid-attn-sample-ratios: 4_2_1
-pyramid-sr-ratios: 2_2_2
-pyramid-embed-dims: 128_256_512
+encoder-embed-dim: 256
+pyramid-stages: 4
+pyramid-layers: 5_5_15_5
+pyramid-sr-ratios: 2_2_2_2
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 256_256_256_256
 pyramid-reduced-embed: conv
 pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1
-pyramid-kernel-sizes: 5_5_5
-pyramid-ffn-ratios: 8_8_4
-pyramid-heads: 2_4_8
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
 
 train-subset: train-clean-100,train-clean-360,train-other-500
 valid-subset: dev-clean
@@ -20,7 +20,7 @@ max-epoch: 100
 max-update: 300000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -38,14 +38,14 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-channels: 1024
 dropout: 0.1
 activation-fn: relu
 encoder-ffn-embed-dim: 2048
-encoder-layers: 6
+encoder-layers: 12
 decoder-layers: 6
 encoder-attention-heads: 4
 

egs/librispeech/asr/conf/pyramid.yaml → egs/librispeech/asr/conf/pyramid4_large.yaml

@@ -2,30 +2,26 @@ arch: pys2t_transformer_s
 encoder-embed-dim: 512
 pyramid-stages: 4
 pyramid-layers: 2_2_6_2
-#encoder-attention-type: reduced
-#pyramid-attn-sample-ratios: 8_4_2_1
-#pyramid-block-attn: True
-#pyramid-fuse-way: add
+#pyramid-layers: 3_3_3_3
 pyramid-sr-ratios: 2_2_2_2
-pyramid-use-ppm: True
-pyramid-embed-dims: 128_128_256_512
-pyramid-reduced-embed: fuse
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 512_512_512_512
+pyramid-reduced-embed: conv
 pyramid-embed-norm: True
 pyramid-position-embed: 1_1_1_1
 pyramid-kernel-sizes: 5_5_5_5
-pyramid-ffn-ratios: 8_8_8_4
-pyramid-heads: 2_2_4_8
-#ctc-layer: 8
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 8_8_8_8
 
-#train-subset: train-clean-100,train-clean-360,train-other-500
-train-subset: train-clean-100
+train-subset: train-clean-100,train-clean-360,train-other-500
 valid-subset: dev-clean
 
 max-epoch: 100
 max-update: 300000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -44,7 +40,6 @@ lr: 2e-3
 #adam_betas: (0.9,0.98)
 
 criterion: label_smoothed_cross_entropy_with_ctc
-ctc-weight: 0.0
 label_smoothing: 0.1
 
 conv-channels: 1024

egs/librispeech/asr/conf/rpr.yaml

-#encoder-attention-type: rel_selfattn
-encoder-attention-type: relative
-max-encoder-relative-length: 100
+encoder-attention-type: rel_selfattn
+#encoder-attention-type: relative
+#max-encoder-relative-length: 100

egs/librispeech/asr/decode.sh

@@ -3,7 +3,7 @@
 gpu_num=1
 
 data_dir=
-test_subset=(test-clean test-other)
+test_subset=(dev-clean dev-other test-clean test-other)
 
 exp_name=
 if [ "$#" -eq 1 ]; then

egs/librispeech/asr/run.sh

@@ -20,7 +20,7 @@ stop_stage=0
 
 ######## hardware ########
 # devices
-#device=()
+device=()
 gpu_num=8
 update_freq=1
 
@@ -251,12 +251,14 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         # Average models
 		dec_model=avg_${n_average}_checkpoint.pt
 
-		cmd="python ${root_dir}/scripts/average_checkpoints.py
-        --inputs ${model_dir}
-        --num-epoch-checkpoints ${n_average}
-        --output ${model_dir}/${dec_model}"
-    	echo -e "\033[34mRun command: \n${cmd} \033[0m"
-    	[[ $eval -eq 1 ]] && eval $cmd
+        if [[ ! -f ${model_dir}/${dec_model} ]]; then
+            cmd="python ${root_dir}/scripts/average_checkpoints.py
+            --inputs ${model_dir}
+            --num-epoch-checkpoints ${n_average}
+            --output ${model_dir}/${dec_model}"
+            echo -e "\033[34mRun command: \n${cmd} \033[0m"
+            [[ $eval -eq 1 ]] && eval $cmd
+        fi
 	else
 		dec_model=${dec_model}
 	fi

egs/librispeech/asr/train.sh

@@ -6,11 +6,14 @@ gpu_num=8
 update_freq=1
 max_tokens=100000
 
-exp_tag=
-config_list=(base)
-config_list=(base conformer)
-#config_list=(pyramid)
-#config_list=(pyramid_stage3)
+#exp_tag=
+#config_list=(base)
+#config_list=(ctc)
+#config_list=(ctc conformer rpr)
+config_list=(base conformer rpr)
+
+#config_list=(pyramid4_all256)
+#config_list=(pyramid5_all256)
 
 # exp full name
 exp_name=
@@ -41,8 +44,7 @@ if [[ -n ${extra_tag} ]]; then
     cmd="$cmd --extra_tag ${extra_tag}"
 fi
 if [[ -n ${extra_parameter} ]]; then
-#    cmd="$cmd --extra_parameter \"${extra_parameter}\""
-    cmd="$cmd --extra_parameter ${extra_parameter}"
+    cmd="$cmd --extra_parameter \"${extra_parameter}\""
 fi
 
 echo ${cmd}

egs/mustc/asr/conf/base.yaml

 train-subset: train_asr
 valid-subset: dev_asr
 
-max-epoch: 50
+max-epoch: 100
 max-update: 100000
 
 num-workers: 8
@@ -24,7 +24,7 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-kernel-sizes: 5,5

egs/mustc/asr/conf/ctc.yaml

-train-subset: train_asr
-valid-subset: dev_asr
-
-max-epoch: 50
-max-update: 100000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
-arch: s2t_transformer_s
-share-decoder-input-output-embed: True
-optimizer: adam
-clip-norm: 10.0
-lr-scheduler: inverse_sqrt
-warmup-init-lr: 1e-7
-warmup-updates: 10000
-lr: 2e-3
-#adam_betas: (0.9,0.98)
-
 ctc-weight: 0.3
-criterion: label_smoothed_cross_entropy_with_ctc
-label_smoothing: 0.1
-
-conv-kernel-sizes: 5,5
-conv-channels: 1024
-dropout: 0.1
-activation-fn: relu
-encoder-embed-dim: 256
-encoder-ffn-embed-dim: 2048
-encoder-layers: 12
-decoder-layers: 6
-encoder-attention-heads: 4
-
-decoder-embed-dim: 256
-decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
-attention-dropout: 0.1
-activation-dropout: 0.1

egs/mustc/asr/conf/pyramid.yaml → egs/mustc/asr/conf/pyramid4_base.yaml

 arch: pys2t_transformer_s
-encoder-embed-dim: 512
+encoder-embed-dim: 256
 pyramid-stages: 4
-pyramid-layers: 2_3_5_2
+#pyramid-dropout: 0
+pyramid-layers: 2_2_6_2
 pyramid-sr-ratios: 2_2_2_2
-pyramid-embed-dims: 128_128_256_512
-pyramid-use-ppm: True
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 256_256_256_256
 pyramid-reduced-embed: conv
 pyramid-embed-norm: True
 pyramid-position-embed: 1_1_1_1
 pyramid-kernel-sizes: 5_5_5_5
-pyramid-ffn-ratios: 8_8_8_4
-pyramid-heads: 2_2_4_8
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
 
 train-subset: train_asr
 valid-subset: dev_asr
 
-max-epoch: 50
+max-epoch: 100
 max-update: 100000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -37,7 +39,7 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-channels: 1024

egs/mustc/asr/conf/pyramid_enc6.yaml → egs/mustc/asr/conf/pyramid4_base_sr8.yaml

 arch: pys2t_transformer_s
-encoder-embed-dim: 512
-pyramid-stages: 3
-pyramid-layers: 2_2_2
-pyramid-sr-ratios: 2_2_2
-pyramid-embed-dims: 128_256_512
+encoder-embed-dim: 256
+pyramid-stages: 4
+#pyramid-dropout: 0
+pyramid-layers: 3_3_3_3
+pyramid-sr-ratios: 2_2_1_2
+pyramid-fuse: True
+pyramid-fuse-way: all_conv
+pyramid-embed-dims: 256_256_256_256
 pyramid-reduced-embed: conv
 pyramid-embed-norm: True
-pyramid-position-embed: 1_0_0
-pyramid-kernel-sizes: 5_5_5
-pyramid-ffn-ratios: 8_8_4
-pyramid-heads: 2_4_8
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
 
 train-subset: train_asr
 valid-subset: dev_asr
 
-max-epoch: 50
+max-epoch: 100
 max-update: 100000
 
 num-workers: 8
-patience: 10
+patience: 20
 no-progress-bar: True
 log-interval: 100
 seed: 1
@@ -36,14 +39,14 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-channels: 1024
 dropout: 0.1
 activation-fn: relu
 encoder-ffn-embed-dim: 2048
-encoder-layers: 6
+encoder-layers: 12
 decoder-layers: 6
 encoder-attention-heads: 4
 

egs/mustc/asr/conf/rpr.yaml

-#encoder-attention-type: rel_selfattn
-encoder-attention-type: relative
-max-encoder-relative-length: 100
+encoder-attention-type: rel_selfattn
+#encoder-attention-type: relative
+#max-encoder-relative-length: 100

egs/mustc/asr/conf/tmp.yaml

-
-

egs/mustc/asr/local/monitor.sh

-gpu_num=1
+gpu_num=4
 cmd="sh train.sh"
 
 while :

egs/mustc/asr/run.sh

@@ -41,8 +41,8 @@ lcrm=0
 tokenizer=0
 
 use_specific_dict=0
-specific_prefix=valid
-specific_dir=/home/xuchen/st/data/mustc/st_lcrm/en-de
+specific_prefix=st
+specific_dir=/home/xuchen/st/data/mustc/st/en-de
 asr_vocab_prefix=spm_unigram10000_st_share
 
 org_data_dir=~/st/data/${dataset}

egs/mustc/asr/train.sh

@@ -6,8 +6,14 @@ gpu_num=8
 update_freq=1
 max_tokens=40000
 
-exp_tag=baseline
-config_list=(base)
+exp_tag=
+
+#config_list=(base)
+#config_list=(ctc)
+#config_list=(base conformer)
+
+#config_list=(pyramid4_base)
+config_list=(pyramid4_base conformer rpr)
 
 # exp full name
 exp_name=

egs/mustc/mt/decode.sh

@@ -13,7 +13,7 @@ fi
 n_average=10
 beam_size=5
 len_penalty=1.0
-max_tokens=10000
+max_tokens=80000
 dec_model=checkpoint_best.pt
 
 cmd="./run.sh
@@ -31,9 +31,9 @@ cmd="./run.sh
 if [[ -n ${data_dir} ]]; then
     cmd="$cmd --data_dir ${data_dir}"
 fi
-if [[ ${#test_subset[@]} -ne 0 ]]; then
-    subsets=$(echo ${test_subset[*]} | sed 's/ /,/g')
-    cmd="$cmd --test_subset ${subsets}"
+if [[ -n ${test_subset} ]]; then
+    test_subset=`echo ${test_subset[*]} | sed 's/ /,/g'`
+    cmd="$cmd --test_subset ${test_subset}"
 fi
 
 echo $cmd

egs/mustc/mt/run.sh

@@ -20,7 +20,7 @@ stop_stage=0
 
 ######## hardware ########
 # devices
-#device=()
+device=()
 gpu_num=8
 update_freq=1
 
@@ -41,10 +41,10 @@ lcrm=0
 tokenizer=0
 
 use_specific_dict=0
-specific_prefix=wmt_share32k
-specific_dir=/home/xuchen/st/data/wmt/mt_lcrm/en-de/unigram32000_share
-src_vocab_prefix=spm_unigram32000_share
-tgt_vocab_prefix=spm_unigram32000_share
+specific_prefix=st
+specific_dir=/home/xuchen/st/data/mustc/st/en-de/
+src_vocab_prefix=spm_unigram10000_st_share
+tgt_vocab_prefix=spm_unigram10000_st_share
 
 org_data_dir=~/st/data/${dataset}
 data_dir=~/st/data/${dataset}/mt/${lang}
@@ -54,14 +54,14 @@ trans_subset=tst-COMMON
 test_subset=test
 
 # exp
-exp_prefix=$(date "+%m%d")
+exp_prefix=${time}
 extra_tag=
 extra_parameter=
 exp_tag=baseline
 exp_name=
 
 # config
-train_config=base
+train_config=base_s
 
 # training setting
 fp16=1
@@ -103,6 +103,7 @@ fi
 
 . ./local/parse_options.sh || exit 1;
 
+# full path
 if [[ -z ${exp_name} ]]; then
     config_string=${train_config//,/_}
 #    exp_name=${exp_prefix}_$(basename ${train_config%.*})_${exp_tag}
@@ -150,7 +151,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     mkdir -p ${data_dir}/data
     for split in ${train_subset} ${valid_subset} ${trans_subset}; do
     {
-        cmd="cat ${org_data_dir}/${lang}/data/${split}.${src_lang}"
+        cmd="cat ${org_data_dir}/${lang}/data/${split}/txt/${split}.${src_lang}"
         if [[ ${lcrm} -eq 1 ]]; then
             cmd="python local/lower_rm.py ${org_data_dir}/${lang}/data/${split}.${src_lang}"
         fi
@@ -165,7 +166,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         cmd="spm_encode
         --model ${data_dir}/${tgt_vocab_prefix}.model
         --output_format=piece
-        < ${org_data_dir}/${lang}/data/${split}.${tgt_lang}
+        < ${org_data_dir}/${lang}/data/${split}/txt/${split}.${tgt_lang}
         > ${data_dir}/data/${split}.${tgt_lang}"
 
         echo -e "\033[34mRun command: \n${cmd} \033[0m"
@@ -178,7 +179,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         --source-lang ${src_lang} --target-lang ${tgt_lang}
         --trainpref ${data_dir}/data/${train_subset}
         --validpref ${data_dir}/data/${valid_subset}
-        --testpref ${data_dir}/data/${test_subset}
+        --testpref ${data_dir}/data/${trans_subset}
         --destdir ${data_dir}/data-bin
         --srcdict ${data_dir}/${src_vocab_prefix}.txt
         --tgtdict ${data_dir}/${tgt_vocab_prefix}.txt
@@ -265,7 +266,7 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
         save_interval=1
         keep_last_epochs=10
         no_epoch_checkpoints=0
-        save_interval_updates=10000
+        save_interval_updates=500
         keep_interval_updates=10
     else
         validate_interval=1
@@ -352,7 +353,7 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 	result_file=${model_dir}/decode_result
 	[[ -f ${result_file} ]] && rm ${result_file}
 
-    test_subset=${test_subset//,/ }
+    test_subset=(${test_subset//,/ })
 	for subset in ${test_subset[@]}; do
   		cmd="python ${root_dir}/fairseq_cli/generate.py
         ${data_dir}

egs/mustc/mt/train.sh

@@ -2,9 +2,9 @@
 
 # training the model
 
-gpu_num=8
+gpu_num=1
 update_freq=1
-max_tokens=4096
+max_tokens=8192
 
 exp_tag=baseline
 config_list=(base)

egs/mustc/st/conf/base.yaml

 train-subset: train_st
 valid-subset: dev_st
 
-max-epoch: 50
+max-epoch: 100
 max-update: 100000
 
 num-workers: 8
@@ -24,7 +24,7 @@ warmup-updates: 10000
 lr: 2e-3
 #adam_betas: (0.9,0.98)
 
-criterion: label_smoothed_cross_entropy
+criterion: label_smoothed_cross_entropy_with_ctc
 label_smoothing: 0.1
 
 conv-kernel-sizes: 5,5

egs/mustc/st/conf/ctc.yaml

-train-subset: train_st
-valid-subset: dev_st
-
-max-epoch: 50
-max-update: 100000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
-arch: s2t_transformer_s
-share-decoder-input-output-embed: True
-optimizer: adam
-clip-norm: 10.0
-lr-scheduler: inverse_sqrt
-warmup-init-lr: 1e-7
-warmup-updates: 10000
-lr: 2e-3
-#adam_betas: (0.9,0.98)
-
-ctc-weight: 0.3
-criterion: label_smoothed_cross_entropy_with_ctc
-label_smoothing: 0.1
-
-conv-kernel-sizes: 5,5
-conv-channels: 1024
-dropout: 0.1
-activation-fn: relu
-encoder-embed-dim: 256
-encoder-ffn-embed-dim: 2048
-encoder-layers: 12
-decoder-layers: 6
-encoder-attention-heads: 4
-
-decoder-embed-dim: 256
-decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
-attention-dropout: 0.1
-activation-dropout: 0.1
+ctc-weight: 0.3
\ No newline at end of file

egs/mustc/st/conf/pyramid4_base_sr8.yaml

+arch: pys2t_transformer_s
+encoder-embed-dim: 256
+#pyramid-dropout: 0
+pyramid-stages: 4
+pyramid-layers: 3_3_3_3
+pyramid-sr-ratios: 2_2_1_2
+pyramid-embed-dims: 256_256_256_256
+pyramid-fuse: True
+pyramid-reduced-embed: conv
+pyramid-embed-norm: True
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
+
+train-subset: train_st
+valid-subset: dev_st
+
+max-epoch: 100
+max-update: 100000
+
+num-workers: 8
+patience: 10
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+
+#load-pretrained-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1002_pyramid4_all256_3333_sr8/avg_10_checkpoint.pt
+#load-pretrained-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1002_pyramid4_all256_3333_sr8/checkpoint_best.pt
+load-pretrained-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1007_st_pyramid4_all256_3333_sr8_ctc/avg_10_checkpoint.pt
+load-pretrained-decoder-from: /home/xuchen/st/checkpoints/mustc/mt/st_1003_2349_train_s_baseline/avg_10_checkpoint.pt
+
+share-decoder-input-output-embed: True
+optimizer: adam
+clip-norm: 10.0
+lr-scheduler: inverse_sqrt
+warmup-init-lr: 1e-7
+warmup-updates: 10000
+lr: 2e-3
+#adam_betas: (0.9,0.98)
+
+criterion: label_smoothed_cross_entropy_with_ctc
+label_smoothing: 0.1
+
+conv-channels: 1024
+dropout: 0.1
+activation-fn: relu
+encoder-ffn-embed-dim: 2048
+encoder-layers: 12
+decoder-layers: 6
+encoder-attention-heads: 4
+
+decoder-embed-dim: 256
+decoder-ffn-embed-dim: 2048
+decoder-attention-heads: 4
+attention-dropout: 0.1
+activation-dropout: 0.1

egs/mustc/st/conf/sate_ctc.yaml

 train-subset: train_st
 valid-subset: dev_st
 
-max-epoch: 50
+max-epoch: 100
 max-update: 100000
 
 num-workers: 8
@@ -50,8 +50,22 @@ cnn-module-kernel: 31
 acoustic-encoder: transformer
 adapter: league
 
-#decoder-embed-dim: 256
-#decoder-ffn-embed-dim: 2048
-#decoder-attention-heads: 4
-#attention-dropout: 0.1
-#activation-dropout: 0.1
+encoder-embed-dim: 256
+pyramid-stages: 4
+#pyramid-dropout: 0
+pyramid-layers: 3_3_3_3
+pyramid-sr-ratios: 2_2_1_2
+pyramid-embed-dims: 256_256_256_256
+pyramid-fuse: True
+pyramid-reduced-embed: conv
+pyramid-embed-norm: True
+pyramid-position-embed: 1_1_1_1
+pyramid-kernel-sizes: 5_5_5_5
+pyramid-ffn-ratios: 8_8_8_8
+pyramid-heads: 4_4_4_4
+
+decoder-embed-dim: 256
+decoder-ffn-embed-dim: 2048
+decoder-attention-heads: 4
+attention-dropout: 0.1
+activation-dropout: 0.1

egs/mustc/st/conf/tmp.yaml

-
-

egs/mustc/st/decode.sh

@@ -13,7 +13,7 @@ fi
 n_average=10
 beam_size=5
 len_penalty=1.0
-max_tokens=10000
+max_tokens=80000
 dec_model=checkpoint_best.pt
 
 cmd="./run.sh

egs/mustc/st/train.sh

@@ -6,8 +6,16 @@ gpu_num=8
 update_freq=1
 max_tokens=40000
 
-exp_tag=baseline
+exp_tag=
+
+#config_list=(base)
 config_list=(ctc)
+#config_list=(sate_ctc)
+#config_list=(ctc conformer rpr)
+#config_list=(base sate)
+
+#config_list=(pyramid4_base_sr8)
+#config_list=(pyramid4_base_sr8 conformer)
 
 # exp full name
 exp_name=