fix the bugs

egs/libri_trans/st/conf/base.yaml → egs/aishell/asr/conf/base.yaml

-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:
-#load-pretrained-decoder-from:
-
 arch: s2t_transformer_s
 arch: s2t_transformer_s
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/aishell/asr/conf/basis.yaml

+train-subset: train
+valid-subset: dev
+
+max-epoch: 100
+max-update: 100000
+patience: 20
+best_checkpoint_metric: loss
+maximize_best_checkpoint_metric: False
+
+no-epoch-checkpoints: True
+#keep-last-epochs: 10
+keep-best-checkpoints: 10
+
+num-workers: 8
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+skip-invalid-size-inputs-valid-test: True
\ No newline at end of file

egs/libri_trans/asr/conf/ctc.yaml → egs/aishell/asr/conf/ctc.yaml

 ctc-weight: 0.3
 ctc-weight: 0.3
+post-process: sentencepiece
\ No newline at end of file

egs/libri_trans/asr/conf/base.yaml → egs/aishell/asr/conf/debug.yaml

@@ -15,6 +15,7 @@ report-accuracy: True
 #load-pretrained-decoder-from:
 #load-pretrained-decoder-from:
 
 
 arch: s2t_transformer_s
 arch: s2t_transformer_s
+#arch: pdss2t_transformer_s
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
 clip-norm: 10.0
 clip-norm: 10.0

egs/libri_trans/asr/conf/pds_base.yaml → egs/aishell/asr/conf/pds_base.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
+pds-fusion: True
+ctc-layer: 12
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/template/st/conf/pds_base.yaml → egs/aishell/asr/conf/pds_base_16.yaml

-arch: pdss2t_transformer_s_8
-
-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:
-#load-pretrained-decoder-from:
+arch: pdss2t_transformer_s_16
+
+encoder-embed-dim: 256
+pds-stages: 4
+ctc-layer: 12
+pds-layers: 2_2_6_2
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/template/asr/conf/pds_base.yaml → egs/aishell/asr/conf/pds_base_32.yaml

-arch: pdss2t_transformer_s_8
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
+arch: pdss2t_transformer_s_32
+
+encoder-embed-dim: 256
+pds-stages: 5
+ctc-layer: 12
+pds-layers: 2_2_3_3_2
+pds-ratios: 2_2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1_1
+pds-kernel-sizes: 5_5_5_5_5
+pds-ffn-ratios: 8_8_8_8_8
+pds-attn-heads: 4_4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/libri_trans/st/conf/pds_base.yaml → egs/aishell/asr/conf/pds_base_8.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
 
 
-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:
-#load-pretrained-decoder-from:
+encoder-embed-dim: 256
+pds-stages: 4
+ctc-layer: 12
+pds-layers: 3_3_3_3
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/libri_trans/asr/decode.sh → egs/aishell/asr/decode.sh

@@ -10,6 +10,7 @@ if [ "$#" -eq 1 ]; then
     exp_name=$1
     exp_name=$1
 fi
 fi
 
 
+cer=1
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
 len_penalty=1.0
 len_penalty=1.0
@@ -21,6 +22,7 @@ cmd="./run.sh
     --stop_stage 2
     --stop_stage 2
     --gpu_num ${gpu_num}
     --gpu_num ${gpu_num}
     --exp_name ${exp_name}
     --exp_name ${exp_name}
+    --cer ${cer}
     --n_average ${n_average}
     --n_average ${n_average}
     --beam_size ${beam_size}
     --beam_size ${beam_size}
     --len_penalty ${len_penalty}
     --len_penalty ${len_penalty}

egs/libri_trans/asr/run.sh → egs/aishell/asr/run.sh

@@ -24,31 +24,31 @@ stop_stage=0
 gpu_num=0
 gpu_num=0
 update_freq=1
 update_freq=1
 
 
-s2t_dir=~/Code/st
-root_dir=${s2t_dir}/Fairseq-S2T
+root_dir=~/st
+code_dir=${root_dir}/Fairseq-S2T
 pwd_dir=$PWD
 pwd_dir=$PWD
 
 
 # dataset
 # dataset
-src_lang=en
-tgt_lang=fr
-lang=${src_lang}-${tgt_lang}
+src_lang=zh
+lang=${src_lang}
 
 
-dataset=libri_trans
+dataset=aishell
 task=speech_to_text
 task=speech_to_text
 vocab_type=unigram
 vocab_type=unigram
-vocab_size=1000
-speed_perturb=0
-lcrm=1
+vocab_type=char
+vocab_size=5000
+speed_perturb=1
+lcrm=0
 tokenizer=0
 tokenizer=0
-use_raw_audio=1
+use_raw_audio=0
 
 
 use_specific_dict=0
 use_specific_dict=0
 specific_prefix=st
 specific_prefix=st
-specific_dir=${s2t_dir}/data/mustc/st/en-de
+specific_dir=${root_dir}/data/mustc/st/en-de
 asr_vocab_prefix=spm_unigram10000_st_share
 asr_vocab_prefix=spm_unigram10000_st_share
 
 
-org_data_dir=${s2t_dir}/data/${dataset}
-data_dir=${s2t_dir}/data/${dataset}/asr
+org_data_dir=${root_dir}/data/${dataset}
+data_dir=${root_dir}/data/${dataset}/asr
 train_split=train
 train_split=train
 valid_split=dev
 valid_split=dev
 test_split=test
 test_split=test
@@ -71,6 +71,7 @@ max_tokens=40000
 step_valid=0
 step_valid=0
 
 
 # decoding setting
 # decoding setting
+cer=1
 dec_model=checkpoint_best.pt
 dec_model=checkpoint_best.pt
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
@@ -96,6 +97,10 @@ if [[ ${use_raw_audio} -eq 1 ]]; then
     data_dir=${data_dir}_raw
     data_dir=${data_dir}_raw
     exp_prefix=${exp_prefix}_raw
     exp_prefix=${exp_prefix}_raw
 fi
 fi
+if [[ "${vocab_type}" == "char" ]]; then
+    data_dir=${data_dir}_char
+    exp_prefix=${exp_prefix}_char
+fi
 
 
 . ./local/parse_options.sh || exit 1;
 . ./local/parse_options.sh || exit 1;
 
 
@@ -106,7 +111,7 @@ if [[ -z ${exp_name} ]]; then
         exp_name=${exp_name}_${extra_tag}
         exp_name=${exp_name}_${extra_tag}
     fi
     fi
 fi
 fi
-model_dir=$root_dir/../checkpoints/$dataset/asr/${exp_name}
+model_dir=$code_dir/../checkpoints/$dataset/asr/${exp_name}
 
 
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
     echo "stage -1: Data Download"
     echo "stage -1: Data Download"
@@ -120,13 +125,19 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     if [[ ! -e ${data_dir} ]]; then
     if [[ ! -e ${data_dir} ]]; then
         mkdir -p ${data_dir}
         mkdir -p ${data_dir}
     fi
     fi
+    if [[ ! -f ${data_dir}/fbank80.zip && -f ${data_dir}/../fbank80.zip ]]; then
+        ln -s ${data_dir}/../fbank80.zip ${data_dir}
+    fi
+    if [[ ! -f ${data_dir}/fbank80_sp.zip && -f ${data_dir}/../fbank80_sp.zip ]]; then
+        ln -s ${data_dir}/../fbank80_sp.zip ${data_dir}
+    fi
 
 
-    cmd="python ${root_dir}/examples/speech_to_text/prep_audio_data.py
+    cmd="python ${code_dir}/examples/speech_to_text/prep_audio_data.py
         --data-root ${org_data_dir}
         --data-root ${org_data_dir}
         --output-root ${data_dir}
         --output-root ${data_dir}
         --task asr
         --task asr
-        --splits ${train_split},${valid_split},${test_split}
         --src-lang ${src_lang}
         --src-lang ${src_lang}
+        --splits ${valid_split},${test_split},${train_split}
         --vocab-type ${vocab_type}
         --vocab-type ${vocab_type}
         --vocab-size ${vocab_size}"
         --vocab-size ${vocab_size}"
 
 
@@ -135,7 +146,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         --raw"
         --raw"
     fi
     fi
     if [[ ${use_specific_dict} -eq 1 ]]; then
     if [[ ${use_specific_dict} -eq 1 ]]; then
-        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}/${lang}
+        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}
         cmd="$cmd
         cmd="$cmd
         --asr-prefix ${asr_vocab_prefix}"
         --asr-prefix ${asr_vocab_prefix}"
     fi
     fi
@@ -155,6 +166,15 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
 
 
     echo -e "\033[34mRun command: \n${cmd} \033[0m"
     echo -e "\033[34mRun command: \n${cmd} \033[0m"
     [[ $eval -eq 1 ]] && eval ${cmd}
     [[ $eval -eq 1 ]] && eval ${cmd}
+
+    if [[ ! -f ${data_dir}/../fbank80.zip ]]; then
+        mv ${data_dir}/fbank80.zip ${data_dir}/..
+        ln -s ${data_dir}/../fbank80.zip ${data_dir}
+    fi
+    if [[ ! -f ${data_dir}/../fbank80_sp.zip ]]; then
+        mv ${data_dir}/fbank80_sp.zip ${data_dir}/..
+        ln -s ${data_dir}/../fbank80_sp.zip ${data_dir}
+    fi
 fi
 fi
 
 
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
@@ -181,28 +201,26 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
 
 
+    extra_parameter="${extra_parameter}
+        --train-config ${pwd_dir}/conf/basis.yaml"
+    cp ${pwd_dir}/conf/basis.yaml ${model_dir}
     config_list="${train_config//,/ }"
     config_list="${train_config//,/ }"
-    idx=0
+    idx=1
     for config in ${config_list[@]}
     for config in ${config_list[@]}
     do
     do
-        config_path=$pwd_dir/conf/${config}.yaml
+        config_path=${pwd_dir}/conf/${config}.yaml
         if [[ ! -f ${config_path} ]]; then
         if [[ ! -f ${config_path} ]]; then
             echo "No config file ${config_path}"
             echo "No config file ${config_path}"
             exit
             exit
         fi
         fi
         cp ${config_path} ${model_dir}
         cp ${config_path} ${model_dir}
 
 
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
+        extra_parameter="${extra_parameter}
         --train-config${idx} ${config_path}"
         --train-config${idx} ${config_path}"
-        fi
         idx=$((idx + 1))
         idx=$((idx + 1))
     done
     done
 
 
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
+    cmd="python3 -u ${code_dir}/fairseq_cli/train.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --task ${task}
         --task ${task}
@@ -286,12 +304,14 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         # Average models
         # Average models
 		dec_model=avg_${n_average}_checkpoint.pt
 		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 ${code_dir}/scripts/average_checkpoints.py
+            --inputs ${model_dir}
+            --num-best-checkpoints ${n_average}
+            --output ${model_dir}/${dec_model}"
+            echo -e "\033[34mRun command: \n${cmd} \033[0m"
+            [[ $eval -eq 1 ]] && eval $cmd
+        fi
 	else
 	else
 		dec_model=${dec_model}
 		dec_model=${dec_model}
 	fi
 	fi
@@ -311,8 +331,8 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 
 
     test_subset=${test_subset//,/ }
     test_subset=${test_subset//,/ }
 	for subset in ${test_subset[@]}; do
 	for subset in ${test_subset[@]}; do
-        subset=${subset}_asr
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
+        subset=${subset}
+  		cmd="python ${code_dir}/fairseq_cli/generate.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --gen-subset ${subset}
         --gen-subset ${subset}
@@ -323,10 +343,13 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         --beam ${beam_size}
         --beam ${beam_size}
         --lenpen ${len_penalty}
         --lenpen ${len_penalty}
         --scoring wer
         --scoring wer
-        --wer-tokenizer 13a
-        --wer-lowercase
-        --wer-remove-punct
         "
         "
+
+        if [[ ${cer} -eq 1 ]]; then
+            cmd="${cmd}
+        --wer-char-level"
+        fi
+
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
 
 
         if [[ $eval -eq 1 ]]; then
         if [[ $eval -eq 1 ]]; then

egs/libri_trans/mt/binary.sh → egs/iwslt14/mt/binary.sh

 set -e
 set -e
 
 
 eval=1
 eval=1
+lcrm=0
 
 
 root_dir=~/st/Fairseq-S2T
 root_dir=~/st/Fairseq-S2T
 data_dir=/home/xuchen/st/data/wmt/test
 data_dir=/home/xuchen/st/data/wmt/test

egs/libri_trans/mt/conf/base.yaml → egs/iwslt14/mt/conf/base.yaml

-train-subset: train
-valid-subset: valid
-
-max-epoch: 50
-max-update: 100000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-skip-invalid-size-inputs-valid-test: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
 arch: transformer
 arch: transformer
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -28,19 +11,22 @@ adam_betas: (0.9,0.997)
 criterion: label_smoothed_cross_entropy
 criterion: label_smoothed_cross_entropy
 label_smoothing: 0.1
 label_smoothing: 0.1
 
 
-dropout: 0.1
-attention-dropout: 0.1
-activation-dropout: 0.1
+dropout: 0.3
+attention-dropout: 0.0
+activation-dropout: 0.0
 
 
 activation-fn: relu
 activation-fn: relu
 encoder-normalize-before: True
 encoder-normalize-before: True
 decoder-normalize-before: True
 decoder-normalize-before: True
 encoder-embed-dim: 512
 encoder-embed-dim: 512
-encoder-ffn-embed-dim: 2048
+encoder-ffn-embed-dim: 1024
 encoder-layers: 6
 encoder-layers: 6
 decoder-layers: 6
 decoder-layers: 6
-encoder-attention-heads: 8
+encoder-attention-heads: 4
 
 
 decoder-embed-dim: 512
 decoder-embed-dim: 512
-decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 8
+decoder-ffn-embed-dim: 1024
+decoder-attention-heads: 4
+
+load-pretrained-encoder-from:
+load-pretrained-decoder-from:
\ No newline at end of file

egs/iwslt14/mt/conf/base_fair.yaml

+arch: transformer_iwslt_de_en
+share-decoder-input-output-embed: True
+optimizer: adam
+#clip-norm: 10.0
+lr-scheduler: inverse_sqrt
+weight-decay: 0.0001
+warmup-init-lr: 1e-7
+warmup-updates: 4000
+lr: 5e-4
+adam_betas: (0.9,0.98)
+
+criterion: label_smoothed_cross_entropy
+label_smoothing: 0.1
+
+dropout: 0.3
+
+activation-fn: relu
+encoder-normalize-before: False
+decoder-normalize-before: False
+encoder-embed-dim: 512
+encoder-ffn-embed-dim: 1024
+encoder-layers: 6
+decoder-layers: 6
+encoder-attention-heads: 4
+
+decoder-embed-dim: 512
+decoder-ffn-embed-dim: 1024
+decoder-attention-heads: 4
+
+load-pretrained-encoder-from:
+load-pretrained-decoder-from:
\ No newline at end of file

egs/libri_trans/mt/conf/base_s.yaml → egs/iwslt14/mt/conf/base_iwslt.yaml

-train-subset: train
-valid-subset: valid
-
-max-epoch: 50
-max-update: 100000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-skip-invalid-size-inputs-valid-test: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
 arch: transformer
 arch: transformer
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -28,19 +11,22 @@ adam_betas: (0.9,0.997)
 criterion: label_smoothed_cross_entropy
 criterion: label_smoothed_cross_entropy
 label_smoothing: 0.1
 label_smoothing: 0.1
 
 
-dropout: 0.1
-attention-dropout: 0.1
-activation-dropout: 0.1
+dropout: 0.3
+attention-dropout: 0.0
+activation-dropout: 0.0
 
 
 activation-fn: relu
 activation-fn: relu
 encoder-normalize-before: True
 encoder-normalize-before: True
 decoder-normalize-before: True
 decoder-normalize-before: True
-encoder-embed-dim: 256
-encoder-ffn-embed-dim: 2048
+encoder-embed-dim: 512
+encoder-ffn-embed-dim: 1024
 encoder-layers: 6
 encoder-layers: 6
 decoder-layers: 6
 decoder-layers: 6
 encoder-attention-heads: 4
 encoder-attention-heads: 4
 
 
-decoder-embed-dim: 256
-decoder-ffn-embed-dim: 2048
+decoder-embed-dim: 512
+decoder-ffn-embed-dim: 1024
 decoder-attention-heads: 4
 decoder-attention-heads: 4
+
+load-pretrained-encoder-from:
+load-pretrained-decoder-from:
\ No newline at end of file

egs/iwslt14/mt/conf/basis.yaml

+train-subset: train
+valid-subset: valid
+
+max-epoch: 50
+max-update: 50000
+patience: 20
+best_checkpoint_metric: loss
+maximize_best_checkpoint_metric: False
+
+no-epoch-checkpoints: True
+#keep-last-epochs: 10
+keep-best-checkpoints: 10
+
+num-workers: 8
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+skip-invalid-size-inputs-valid-test: True
\ No newline at end of file

egs/libri_trans/mt/conf/rpr.yaml → egs/iwslt14/mt/conf/rpr.yaml

@@ -2,4 +2,4 @@
 encoder-attention-type: relative
 encoder-attention-type: relative
 decoder-attention-type: relative
 decoder-attention-type: relative
 max-encoder-relative-length: 20
 max-encoder-relative-length: 20
-max-decoder-relative-length: 20
+max-decoder-relative-length: 20
\ No newline at end of file

egs/libri_trans/mt/decode.sh → egs/iwslt14/mt/decode.sh

@@ -10,6 +10,7 @@ if [ "$#" -eq 1 ]; then
     exp_name=$1
     exp_name=$1
 fi
 fi
 
 
+sacrebleu=0
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
 len_penalty=1.0
 len_penalty=1.0
@@ -21,6 +22,7 @@ cmd="./run.sh
     --stop_stage 2
     --stop_stage 2
     --gpu_num ${gpu_num}
     --gpu_num ${gpu_num}
     --exp_name ${exp_name}
     --exp_name ${exp_name}
+    --sacrebleu ${sacrebleu}
     --n_average ${n_average}
     --n_average ${n_average}
     --beam_size ${beam_size}
     --beam_size ${beam_size}
     --len_penalty ${len_penalty}
     --len_penalty ${len_penalty}

egs/libri_trans/mt/run.sh → egs/iwslt14/mt/run.sh

@@ -13,7 +13,7 @@ set -o pipefail
 export PYTHONIOENCODING=UTF-8
 export PYTHONIOENCODING=UTF-8
 
 
 eval=1
 eval=1
-time=$(date "+%m%d_%H%M")
+time=$(date "+%m%d")
 
 
 stage=0
 stage=0
 stop_stage=0
 stop_stage=0
@@ -24,33 +24,34 @@ device=()
 gpu_num=8
 gpu_num=8
 update_freq=1
 update_freq=1
 
 
-root_dir=~/st/Fairseq-S2T
+root_dir=~/st
+code_dir=${root_dir}/Fairseq-S2T
 pwd_dir=$PWD
 pwd_dir=$PWD
 
 
 # dataset
 # dataset
-src_lang=en
-tgt_lang=de
+src_lang=de
+tgt_lang=en
 lang=${src_lang}-${tgt_lang}
 lang=${src_lang}-${tgt_lang}
 
 
-dataset=mt
+dataset=iwslt14
 task=translation
 task=translation
 vocab_type=unigram
 vocab_type=unigram
 vocab_size=10000
 vocab_size=10000
 share_dict=1
 share_dict=1
 lcrm=0
 lcrm=0
-tokenizer=0
+tokenizer=1
 
 
 use_specific_dict=0
 use_specific_dict=0
 specific_prefix=st
 specific_prefix=st
-specific_dir=/home/xuchen/st/data/mustc/st/en-de/
+specific_dir=${root_dir}/data/mustc/st/en-de/
 src_vocab_prefix=spm_unigram10000_st_share
 src_vocab_prefix=spm_unigram10000_st_share
 tgt_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/${lang}
+org_data_dir=${root_dir}/data/${dataset}
+data_dir=${root_dir}/data/${dataset}/mt
 train_subset=train
 train_subset=train
 valid_subset=dev
 valid_subset=dev
-trans_subset=tst-COMMON
+trans_subset=test
 test_subset=test
 test_subset=test
 
 
 # exp
 # exp
@@ -70,6 +71,7 @@ step_valid=0
 bleu_valid=0
 bleu_valid=0
 
 
 # decoding setting
 # decoding setting
+sacrebleu=0
 dec_model=checkpoint_best.pt
 dec_model=checkpoint_best.pt
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
@@ -80,13 +82,19 @@ if [[ ${use_specific_dict} -eq 1 ]]; then
     data_dir=${data_dir}/${specific_prefix}
     data_dir=${data_dir}/${specific_prefix}
     mkdir -p ${data_dir}
     mkdir -p ${data_dir}
 else
 else
-    data_dir=${data_dir}/${vocab_type}${vocab_size}
-    src_vocab_prefix=spm_${vocab_type}${vocab_size}_${src_lang}
-    tgt_vocab_prefix=spm_${vocab_type}${vocab_size}_${tgt_lang}
+    if [[ "${vocab_type}" == "char" ]]; then
+        vocab_name=${vocab_type}
+        exp_prefix=${exp_prefix}_${vocab_type}
+    else
+        vocab_name=${vocab_type}${vocab_size}
+    fi
+    data_dir=${data_dir}/${vocab_name}
+    src_vocab_prefix=spm_${vocab_name}_${src_lang}
+    tgt_vocab_prefix=spm_${vocab_name}_${tgt_lang}
     if [[ $share_dict -eq 1 ]]; then
     if [[ $share_dict -eq 1 ]]; then
         data_dir=${data_dir}_share
         data_dir=${data_dir}_share
-        src_vocab_prefix=spm_${vocab_type}${vocab_size}_share
-        tgt_vocab_prefix=spm_${vocab_type}${vocab_size}_share
+        src_vocab_prefix=spm_${vocab_name}_share
+        tgt_vocab_prefix=spm_${vocab_name}_share
     fi
     fi
 fi
 fi
 if [[ ${lcrm} -eq 1 ]]; then
 if [[ ${lcrm} -eq 1 ]]; then
@@ -111,7 +119,7 @@ if [[ -z ${exp_name} ]]; then
         exp_name=${exp_name}_${extra_tag}
         exp_name=${exp_name}_${extra_tag}
     fi
     fi
 fi
 fi
-model_dir=$root_dir/../checkpoints/$dataset/mt/${exp_name}
+model_dir=$code_dir/../checkpoints/$dataset/mt/${exp_name}
 
 
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
     echo "stage -1: Data Download"
     echo "stage -1: Data Download"
@@ -127,7 +135,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
 
 
     if [[ ! -f ${data_dir}/${src_vocab_prefix}.txt || ! -f ${data_dir}/${tgt_vocab_prefix}.txt ]]; then
     if [[ ! -f ${data_dir}/${src_vocab_prefix}.txt || ! -f ${data_dir}/${tgt_vocab_prefix}.txt ]]; then
         if [[ ${use_specific_dict} -eq 0 ]]; then
         if [[ ${use_specific_dict} -eq 0 ]]; then
-            cmd="python ${root_dir}/examples/speech_to_text/prep_mt_data.py
+            cmd="python ${code_dir}/examples/speech_to_text/prep_mt_data.py
                 --data-root ${org_data_dir}
                 --data-root ${org_data_dir}
                 --output-root ${data_dir}
                 --output-root ${data_dir}
                 --splits ${train_subset},${valid_subset},${trans_subset}
                 --splits ${train_subset},${valid_subset},${trans_subset}
@@ -150,9 +158,14 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     mkdir -p ${data_dir}/data
     mkdir -p ${data_dir}/data
     for split in ${train_subset} ${valid_subset} ${trans_subset}; do
     for split in ${train_subset} ${valid_subset} ${trans_subset}; do
     {
     {
-        cmd="cat ${org_data_dir}/${lang}/data/${split}/txt/${split}.${src_lang}"
+        if [[ -d ${org_data_dir}/data/${split}/txt ]]; then
+            txt_dir=${org_data_dir}/data/${split}/txt
+        else
+            txt_dir=${org_data_dir}/data/${split}
+        fi
+        cmd="cat ${txt_dir}/${split}.${src_lang}"
         if [[ ${lcrm} -eq 1 ]]; then
         if [[ ${lcrm} -eq 1 ]]; then
-            cmd="python local/lower_rm.py ${org_data_dir}/${lang}/data/${split}.${src_lang}"
+            cmd="python local/lower_rm.py ${org_data_dir}/data/${split}.${src_lang}"
         fi
         fi
         cmd="${cmd}
         cmd="${cmd}
         | spm_encode --model ${data_dir}/${src_vocab_prefix}.model
         | spm_encode --model ${data_dir}/${src_vocab_prefix}.model
@@ -165,7 +178,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         cmd="spm_encode
         cmd="spm_encode
         --model ${data_dir}/${tgt_vocab_prefix}.model
         --model ${data_dir}/${tgt_vocab_prefix}.model
         --output_format=piece
         --output_format=piece
-        < ${org_data_dir}/${lang}/data/${split}.${tgt_lang}
+        < ${txt_dir}/${split}.${tgt_lang}
         > ${data_dir}/data/${split}.${tgt_lang}"
         > ${data_dir}/data/${split}.${tgt_lang}"
 
 
         echo -e "\033[34mRun command: \n${cmd} \033[0m"
         echo -e "\033[34mRun command: \n${cmd} \033[0m"
@@ -174,7 +187,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     done
     done
     wait
     wait
 
 
-    cmd="python ${root_dir}/fairseq_cli/preprocess.py
+    cmd="python ${code_dir}/fairseq_cli/preprocess.py
         --source-lang ${src_lang} --target-lang ${tgt_lang}
         --source-lang ${src_lang} --target-lang ${tgt_lang}
         --trainpref ${data_dir}/data/${train_subset}
         --trainpref ${data_dir}/data/${train_subset}
         --validpref ${data_dir}/data/${valid_subset}
         --validpref ${data_dir}/data/${valid_subset}
@@ -214,28 +227,26 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
 
 
+    extra_parameter="${extra_parameter}
+        --train-config ${pwd_dir}/conf/basis.yaml"
+    cp ${pwd_dir}/conf/basis.yaml ${model_dir}
     config_list="${train_config//,/ }"
     config_list="${train_config//,/ }"
-    idx=0
+    idx=1
     for config in ${config_list[@]}
     for config in ${config_list[@]}
     do
     do
-        config_path=$pwd_dir/conf/${config}.yaml
+        config_path=${pwd_dir}/conf/${config}.yaml
         if [[ ! -f ${config_path} ]]; then
         if [[ ! -f ${config_path} ]]; then
             echo "No config file ${config_path}"
             echo "No config file ${config_path}"
             exit
             exit
         fi
         fi
         cp ${config_path} ${model_dir}
         cp ${config_path} ${model_dir}
 
 
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
+        extra_parameter="${extra_parameter}
         --train-config${idx} ${config_path}"
         --train-config${idx} ${config_path}"
-        fi
         idx=$((idx + 1))
         idx=$((idx + 1))
     done
     done
 
 
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
+    cmd="python3 -u ${code_dir}/fairseq_cli/train.py
         ${data_dir}
         ${data_dir}
         --source-lang ${src_lang}
         --source-lang ${src_lang}
         --target-lang ${tgt_lang}
         --target-lang ${tgt_lang}
@@ -263,13 +274,9 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     if [[ $step_valid -eq 1 ]]; then
     if [[ $step_valid -eq 1 ]]; then
         validate_interval=1
         validate_interval=1
         save_interval=1
         save_interval=1
-        keep_last_epochs=10
         no_epoch_checkpoints=0
         no_epoch_checkpoints=0
         save_interval_updates=500
         save_interval_updates=500
         keep_interval_updates=10
         keep_interval_updates=10
-    else
-        validate_interval=1
-        keep_last_epochs=10
     fi
     fi
     if [[ $bleu_valid -eq 1 ]]; then
     if [[ $bleu_valid -eq 1 ]]; then
         cmd="$cmd
         cmd="$cmd
@@ -292,10 +299,6 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
         cmd="${cmd}
         cmd="${cmd}
         --save-interval $save_interval "
         --save-interval $save_interval "
     fi
     fi
-    if [[ -n $keep_last_epochs ]]; then
-        cmd="${cmd}
-        --keep-last-epochs $keep_last_epochs "
-    fi
     if [[ -n $save_interval_updates ]]; then
     if [[ -n $save_interval_updates ]]; then
         cmd="${cmd}
         cmd="${cmd}
         --save-interval-updates $save_interval_updates"
         --save-interval-updates $save_interval_updates"
@@ -329,9 +332,9 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         # Average models
         # Average models
 		dec_model=avg_${n_average}_checkpoint.pt
 		dec_model=avg_${n_average}_checkpoint.pt
 
 
-		cmd="python ${root_dir}/scripts/average_checkpoints.py
+		cmd="python ${code_dir}/scripts/average_checkpoints.py
         --inputs ${model_dir}
         --inputs ${model_dir}
-        --num-epoch-checkpoints ${n_average}
+        --num-best-checkpoints ${n_average}
         --output ${model_dir}/${dec_model}"
         --output ${model_dir}/${dec_model}"
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
     	[[ $eval -eq 1 ]] && eval $cmd
     	[[ $eval -eq 1 ]] && eval $cmd
@@ -354,7 +357,7 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 
 
     test_subset=(${test_subset//,/ })
     test_subset=(${test_subset//,/ })
 	for subset in ${test_subset[@]}; do
 	for subset in ${test_subset[@]}; do
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
+  		cmd="python ${code_dir}/fairseq_cli/generate.py
         ${data_dir}
         ${data_dir}
         --source-lang ${src_lang}
         --source-lang ${src_lang}
         --target-lang ${tgt_lang}
         --target-lang ${tgt_lang}
@@ -365,14 +368,17 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         --max-tokens ${max_tokens}
         --max-tokens ${max_tokens}
         --beam ${beam_size}
         --beam ${beam_size}
         --lenpen ${len_penalty}
         --lenpen ${len_penalty}
-        --post-process sentencepiece
-        --scoring sacrebleu"
+        --post-process sentencepiece"
 
 
-        if [[ ${tokenizer} -eq 1 ]]; then
+        if [[ ${sacrebleu} -eq 1 ]]; then
             cmd="${cmd}
             cmd="${cmd}
+        --scoring sacrebleu"
+            if [[ ${tokenizer} -eq 1 ]]; then
+                cmd="${cmd}
         --tokenizer moses
         --tokenizer moses
         --moses-source-lang ${src_lang}
         --moses-source-lang ${src_lang}
         --moses-target-lang ${tgt_lang}"
         --moses-target-lang ${tgt_lang}"
+            fi
         fi
         fi
 
 
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"

egs/libri_trans/asr/conf/pds_base_16.yaml

-arch: pdss2t_transformer_s_16
-
-encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#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
-
-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

egs/libri_trans/asr/conf/pds_base_32.yaml

-arch: pdss2t_transformer_s_32
-
-encoder-embed-dim: 256
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 2_2_3_3_2
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256_256
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 8_8_8_8_8
-pyramid-attn-heads: 4_4_4_4_4
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#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
-
-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

egs/libri_trans/asr/conf/pds_base_8.yaml

-arch: pdss2t_transformer_s_8
-
-encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#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
-
-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

egs/libri_trans/st/binary.sh

-set -e
-
-eval=1
-
-lcrm=1
-tokenizer=0
-
-root_dir=~/st/Fairseq-S2T
-data_dir=/home/xuchen/st/data/test
-vocab_dir=/home/xuchen/st/data/mustc/st_lcrm/en-de
-asr_vocab_prefix=spm_unigram10000_st_share
-st_vocab_prefix=spm_unigram10000_st_share
-
-src_lang=en
-tgt_lang=de
-splits=(2019)
-
-splits=$(echo ${splits[*]} | sed 's/ /_/g')
-
-cp -r ${vocab_dir}/${asr_vocab_prefix}.* ${data_dir}/${src_lang}-${tgt_lang}
-cp -r ${vocab_dir}/${st_vocab_prefix}.* ${data_dir}/${src_lang}-${tgt_lang}
-rm -rf ${data_dir}/${src_lang}-${tgt_lang}/fbank80.zip
-
-cmd="python ${root_dir}/examples/speech_to_text/prep_st_data.py
-    --data-root ${data_dir}
-    --output-root ${data_dir}
-    --splits ${splits}
-    --task st
-    --src-lang ${src_lang}
-    --tgt-lang ${tgt_lang}
-    --add-src
-    --share
-    --asr-prefix ${asr_vocab_prefix}
-    --st-spm-prefix ${st_vocab_prefix}
-    --cmvn-type utterance"
-
-    if [[ ${lcrm} -eq 1 ]]; then
-        cmd="$cmd
-    --lowercase-src
-    --rm-punc-src"
-    fi
-    if [[ ${tokenizer} -eq 1 ]]; then
-        cmd="$cmd
-    --tokenizer"
-    fi
-
-echo -e "\033[34mRun command: \n${cmd} \033[0m"
-[[ $eval -eq 1 ]] && eval ${cmd}

egs/libri_trans/st/conf/conformer.yaml

-macaron-style: True
-use-cnn-module: True
-cnn-module-kernel: 31

egs/libri_trans/st/conf/ctc.yaml

-ctc-weight: 0.3
\ No newline at end of file

egs/libri_trans/st/conf/dlcl.yaml

-use-enc-dlcl: True
-use-dec-dlcl: True

egs/libri_trans/st/conf/local_attn.yaml

-encoder-attention-type: local
-hard-mask-window: 0
-gauss-mask-sigma: 3
-init-mask-weight: 0
\ No newline at end of file

egs/libri_trans/st/conf/pds_base_16.yaml

-arch: pdss2t_transformer_s_16
-
-encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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:
-#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
-
-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

egs/libri_trans/st/conf/pds_base_32.yaml

-arch: pdss2t_transformer_s_32
-
-encoder-embed-dim: 256
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 2_2_3_3_2
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256_256
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 8_8_8_8_8
-pyramid-attn-heads: 4_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:
-#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
-
-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

egs/libri_trans/st/conf/pds_base_8.yaml

-arch: pdss2t_transformer_s_8
-
-encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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:
-#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
-
-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

egs/libri_trans/st/conf/rpr.yaml

-encoder-attention-type: rel_selfattn
-#encoder-attention-type: relative
-#decoder-attention-type: relative
-#max-encoder-relative-length: 100
-#max-decoder-relative-length: 20

egs/libri_trans/st/conf/sate_ctc.yaml

-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:
-#load-pretrained-acoustic-encoder-from:
-#load-pretrained-text-encoder-from:
-#load-pretrained-decoder-from:
-
-arch: s2t_sate
-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
-
-encoder-normalize-before: True
-decoder-normalize-before: True
-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
-text-encoder-layers: 6
-decoder-layers: 6
-encoder-attention-heads: 4
-
-#macaron-style: True
-#use-cnn-module: True
-#cnn-module-kernel: 31
-
-#acoustic-encoder: pds
-acoustic-encoder: transformer
-adapter: league
-
-encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
-
-decoder-embed-dim: 256
-decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4

egs/libri_trans/st/decode.sh

-#! /bin/bash
-
-gpu_num=1
-
-data_dir=
-test_subset=(tst-COMMON)
-
-exp_name=
-if [ "$#" -eq 1 ]; then
-    exp_name=$1
-fi
-
-n_average=10
-beam_size=5
-len_penalty=1.0
-max_tokens=80000
-dec_model=checkpoint_best.pt
-
-cmd="./run.sh
-    --stage 2
-    --stop_stage 2
-    --gpu_num ${gpu_num}
-    --exp_name ${exp_name}
-    --n_average ${n_average}
-    --beam_size ${beam_size}
-    --len_penalty ${len_penalty}
-    --max_tokens ${max_tokens}
-    --dec_model ${dec_model}
-    "
-
-if [[ -n ${data_dir} ]]; then
-    cmd="$cmd --data_dir ${data_dir}"
-fi
-if [[ ${#test_subset[@]} -eq 0 ]]; then
-    subsets=$(echo ${test_subset[*]} | sed 's/ /,/g')
-    cmd="$cmd --test_subset ${subsets}"
-fi
-
-echo $cmd
-eval $cmd

egs/libri_trans/st/ensemble.sh

-set -e
-
-gpu_num=1
-root_dir=/home/xuchen/st/Fairseq-S2T
-ckpt=/home/xuchen/st/checkpoints/mustc-v2/st
-
-model_txt=$1
-set=$2
-test_subset=$3
-
-#data_dir=/home/xuchen/st/data/mustc-v2/st_lcrm/en-de
-#test_subset=(tst-COMMON)
-
-data_dir=/media/data/tst/$set/en-de
-#test_subset=(office)
-#test_subset=(webrtc1)
-#test_subset=(adap2)
-
-data_config=config_st_share.yaml
-result_file=./result
-
-beam_size=5
-lenpen=0.6
-max_tokens=10000
-
-models=()
-i=0
-for line in `cat $model_txt`; do
-    i=`expr $i + 1`
-    
-    model_dir=$ckpt/$line
-    [[ ! -d $model_dir ]] && echo $model_dir && exit 1;
-
-    if [[ -f $model_dir/avg_10_checkpoint.pt ]]; then
-        model=$model_dir/avg_10_checkpoint.pt
-    else
-        model=$model_dir/checkpoint_best.pt
-    fi
-    [[ ! -f $model ]] && echo $model && exit 1;
-
-    models[$i]=$model
-done
-
-models=`echo ${models[*]} | sed 's/ /:/g'`
-
-res_dir=$ckpt/ensemble/$set
-i=0
-while : 
-do
-    if [[ -d $res_dir/$i ]]; then
-        i=`expr $i + 1`
-    else
-        res_dir=$res_dir/$i
-        break
-    fi 
-done
-
-mkdir -p $res_dir
-cp $model_txt $res_dir
-
-
-if [[ -z ${device} || ${#device[@]} -eq 0 ]]; then
-    if [[ ${gpu_num} -eq 0 ]]; then
-        device=()
-    else
-        source ./local/utils.sh
-        device=$(get_devices $gpu_num 0)
-    fi
-fi
-export CUDA_VISIBLE_DEVICES=${device}
-
-for subset in ${test_subset[@]}; do
-    subset=${subset}_st
-    cmd="python ${root_dir}/fairseq_cli/generate.py
-    ${data_dir}
-    --config-yaml ${data_config}
-    --gen-subset ${subset}
-    --task speech_to_text
-    --path ${models}
-    --results-path ${res_dir}
-    --skip-invalid-size-inputs-valid-test
-    --max-tokens ${max_tokens}
-    --beam ${beam_size}
-    --lenpen ${lenpen}
-    --scoring sacrebleu"
-    echo -e "\033[34mRun command: \n${cmd} \033[0m"
-
-    eval $cmd
-    tail -n 1 ${res_dir}/generate-${subset}.txt
-
-    cd $res_dir
-    evaluate.sh translation-${subset}.txt $set
-    cd -
-done
-

egs/libri_trans/st/local/monitor.sh

-gpu_num=4
-cmd="sh train.sh"
-
-while :
-do
-    record=$(mktemp -t temp.record.XXXXXX)
-    gpustat > $record
-    all_devices=$(seq 0 "$(sed '1,2d' ${record} | wc -l)");
-
-    count=0
-    for dev in ${all_devices[@]}
-    do
-        line=$((dev + 2))
-        use=$(head -n $line ${record} | tail -1 | cut -d '|' -f3 | cut -d '/' -f1)
-
-        if [[ $use -lt 100 ]]; then
-            device[$count]=$dev
-            count=$((count + 1))
-            if [[ $count -eq $gpu_num ]]; then
-                break
-            fi
-        fi
-    done
-    if [[ ${#device[@]} -lt $gpu_num ]]; then
-        sleep 60s
-    else
-        echo "Run $cmd"
-        eval $cmd
-        sleep 10s
-        exit
-    fi
-done

egs/libri_trans/st/local/parse_options.sh

-#!/usr/bin/env bash
-
-# Copyright 2012  Johns Hopkins University (Author: Daniel Povey);
-#                 Arnab Ghoshal, Karel Vesely
-
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#  http://www.apache.org/licenses/LICENSE-2.0
-#
-# THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
-# KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
-# WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
-# MERCHANTABLITY OR NON-INFRINGEMENT.
-# See the Apache 2 License for the specific language governing permissions and
-# limitations under the License.
-
-
-# Parse command-line options.
-# To be sourced by another script (as in ". parse_options.sh").
-# Option format is: --option-name arg
-# and shell variable "option_name" gets set to value "arg."
-# The exception is --help, which takes no arguments, but prints the
-# $help_message variable (if defined).
-
-
-###
-### The --config file options have lower priority to command line
-### options, so we need to import them first...
-###
-
-# Now import all the configs specified by command-line, in left-to-right order
-for ((argpos=1; argpos<$#; argpos++)); do
-  if [ "${!argpos}" == "--config" ]; then
-    argpos_plus1=$((argpos+1))
-    config=${!argpos_plus1}
-    [ ! -r $config ] && echo "$0: missing config '$config'" && exit 1
-    . $config  # source the config file.
-  fi
-done
-
-
-###
-### Now we process the command line options
-###
-while true; do
-  [ -z "${1:-}" ] && break;  # break if there are no arguments
-  case "$1" in
-    # If the enclosing script is called with --help option, print the help
-    # message and exit.  Scripts should put help messages in $help_message
-    --help|-h) if [ -z "$help_message" ]; then echo "No help found." 1>&2;
-      else printf "$help_message\n" 1>&2 ; fi;
-      exit 0 ;;
-    --*=*) echo "$0: options to scripts must be of the form --name value, got '$1'"
-      exit 1 ;;
-    # If the first command-line argument begins with "--" (e.g. --foo-bar),
-    # then work out the variable name as $name, which will equal "foo_bar".
-    --*) name=`echo "$1" | sed s/^--// | sed s/-/_/g`;
-      # Next we test whether the variable in question is undefned-- if so it's
-      # an invalid option and we die.  Note: $0 evaluates to the name of the
-      # enclosing script.
-      # The test [ -z ${foo_bar+xxx} ] will return true if the variable foo_bar
-      # is undefined.  We then have to wrap this test inside "eval" because
-      # foo_bar is itself inside a variable ($name).
-      eval '[ -z "${'$name'+xxx}" ]' && echo "$0: invalid option $1" 1>&2 && exit 1;
-
-      oldval="`eval echo \\$$name`";
-      # Work out whether we seem to be expecting a Boolean argument.
-      if [ "$oldval" == "true" ] || [ "$oldval" == "false" ]; then
-        was_bool=true;
-      else
-        was_bool=false;
-      fi
-
-      # Set the variable to the right value-- the escaped quotes make it work if
-      # the option had spaces, like --cmd "queue.pl -sync y"
-      eval $name=\"$2\";
-
-      # Check that Boolean-valued arguments are really Boolean.
-      if $was_bool && [[ "$2" != "true" && "$2" != "false" ]]; then
-        echo "$0: expected \"true\" or \"false\": $1 $2" 1>&2
-        exit 1;
-      fi
-      shift 2;
-      ;;
-  *) break;
-  esac
-done
-
-
-# Check for an empty argument to the --cmd option, which can easily occur as a
-# result of scripting errors.
-[ ! -z "${cmd+xxx}" ] && [ -z "$cmd" ] && echo "$0: empty argument to --cmd option" 1>&2 && exit 1;
-
-
-true; # so this script returns exit code 0.

egs/libri_trans/st/local/utils.sh

-
-get_devices(){
-    gpu_num=$1
-    use_cpu=$2
-    device=()
-    while :
-    do
-        record=$(mktemp -t temp.record.XXXXXX)
-        gpustat > $record
-        all_devices=$(seq 0 "$(sed '1,2d' ${record} | wc -l)");
-
-        count=0
-        for dev in ${all_devices[@]}
-        do
-            line=$((dev + 2))
-            use=$(head -n $line ${record} | tail -1 | cut -d '|' -f3 | cut -d '/' -f1)
-            if [[ $use -lt 100 ]]; then
-                device[$count]=$dev
-                count=$((count + 1))
-                if [[ $count -eq $gpu_num ]]; then
-                    break
-                fi
-            fi
-        done
-        if [[ ${#device[@]} -lt $gpu_num ]]; then
-            if [[ $use_cpu -eq 1 ]]; then
-                device=(-1)
-            else
-                sleep 60s
-            fi
-        else
-            break
-        fi
-    done
-
-    echo ${device[*]} | sed 's/ /,/g'
-    return $?
-}
-
-

egs/libri_trans/st/run.sh

-#! /bin/bash
-
-# Processing MuST-C Datasets
-
-# Copyright 2021 Natural Language Processing Laboratory 
-# Xu Chen (xuchenneu@163.com)
-
-# Set bash to 'debug' mode, it will exit on :
-# -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands',
-set -e
-#set -u
-set -o pipefail
-export PYTHONIOENCODING=UTF-8
-
-eval=1
-time=$(date "+%m%d_%H%M")
-
-stage=0
-stop_stage=0
-
-######## hardware ########
-# devices
-#device=()
-gpu_num=8
-update_freq=1
-
-root_dir=~/st/Fairseq-S2T
-pwd_dir=$PWD
-
-# dataset
-src_lang=en
-tgt_lang=de
-lang=${src_lang}-${tgt_lang}
-
-dataset=st
-task=speech_to_text
-vocab_type=unigram
-asr_vocab_size=5000
-vocab_size=10000
-share_dict=1
-speed_perturb=0
-lcrm=0
-tokenizer=0
-
-use_specific_dict=0
-specific_prefix=valid
-specific_dir=/home/xuchen/st/data/mustc/st_lcrm/en-de
-asr_vocab_prefix=spm_unigram10000_st_share
-st_vocab_prefix=spm_unigram10000_st_share
-
-org_data_dir=~/st/data/${dataset}
-data_dir=~/st/data/${dataset}/st
-test_subset=tst-COMMON
-
-# exp
-exp_prefix=$(date "+%m%d")
-extra_tag=
-extra_parameter=
-exp_tag=baseline
-exp_name=
-
-# config
-train_config=ctc
-
-# training setting
-fp16=1
-max_tokens=40000
-step_valid=0
-bleu_valid=0
-
-# decoding setting
-dec_model=checkpoint_best.pt
-n_average=10
-beam_size=5
-len_penalty=1.0
-
-if [[ ${share_dict} -eq 1 ]]; then
-	data_config=config_st_share.yaml
-else
-	data_config=config_st.yaml
-fi
-if [[ ${speed_perturb} -eq 1 ]]; then
-    data_dir=${data_dir}_sp
-    exp_prefix=${exp_prefix}_sp
-fi
-if [[ ${lcrm} -eq 1 ]]; then
-    data_dir=${data_dir}_lcrm
-    exp_prefix=${exp_prefix}_lcrm
-fi
-if [[ ${use_specific_dict} -eq 1 ]]; then
-    data_dir=${data_dir}_${specific_prefix}
-    exp_prefix=${exp_prefix}_${specific_prefix}
-fi
-if [[ ${tokenizer} -eq 1 ]]; then
-    data_dir=${data_dir}_tok
-    exp_prefix=${exp_prefix}_tok
-fi
-
-. ./local/parse_options.sh || exit 1;
-
-if [[ -z ${exp_name} ]]; then
-    config_string=${train_config//,/_}
-    exp_name=${exp_prefix}_${config_string}_${exp_tag}
-    if [[ -n ${extra_tag} ]]; then
-        exp_name=${exp_name}_${extra_tag}
-    fi
-fi
-model_dir=$root_dir/../checkpoints/$dataset/st/${exp_name}
-
-if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
-    echo "stage -1: Data Download"
-    # pass
-fi
-
-if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
-    ### Task dependent. You have to make data the following preparation part by yourself.
-    ### But you can utilize Kaldi recipes in most cases
-    echo "stage 0: ASR Data Preparation"
-    if [[ ! -e ${data_dir}/${lang} ]]; then
-        mkdir -p ${data_dir}/${lang}
-    fi
-
-    cmd="python ${root_dir}/examples/speech_to_text/prep_asr_data.py
-        --data-root ${org_data_dir}
-        --output-root ${data_dir}
-        --task asr
-        --vocab-type ${vocab_type}
-        --vocab-size ${asr_vocab_size}"
-    if [[ ${speed_perturb} -eq 1 ]]; then
-        cmd="$cmd
-        --speed-perturb"
-    fi
-    echo -e "\033[34mRun command: \n${cmd} \033[0m"
-    [[ $eval -eq 1 && ${share_dict} -ne 1 && ${use_specific_dict} -ne 1 ]] && eval $cmd
-    asr_prefix=spm_${vocab_type}${asr_vocab_size}_asr
-
-    echo "stage 0: ST Data Preparation"
-    cmd="python ${root_dir}/examples/speech_to_text/prep_st_data.py
-        --data-root ${org_data_dir}
-        --output-root ${data_dir}
-        --task st
-        --add-src
-        --cmvn-type utterance
-        --vocab-type ${vocab_type}
-        --vocab-size ${vocab_size}"
-
-    if [[ ${use_specific_dict} -eq 1 ]]; then
-        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}/${lang}
-        cp -r ${specific_dir}/${st_vocab_prefix}.* ${data_dir}/${lang}
-        if [[ $share_dict -eq 1 ]]; then
-            cmd="$cmd
-        --share
-        --st-spm-prefix ${st_vocab_prefix}"
-        else
-            cmd="$cmd
-        --st-spm-prefix ${st_vocab_prefix}
-        --asr-prefix ${asr_vocab_prefix}"
-        fi
-    else
-        if [[ $share_dict -eq 1 ]]; then
-            cmd="$cmd
-        --share"
-        else
-            cmd="$cmd
-        --asr-prefix ${asr_prefix}"
-        fi
-    fi
-    if [[ ${speed_perturb} -eq 1 ]]; then
-        cmd="$cmd
-        --speed-perturb"
-    fi
-    if [[ ${lcrm} -eq 1 ]]; then
-        cmd="$cmd
-        --lowercase-src
-        --rm-punc-src"
-    fi
-    if [[ ${tokenizer} -eq 1 ]]; then
-        cmd="$cmd
-        --tokenizer"
-    fi
-
-    echo -e "\033[34mRun command: \n${cmd} \033[0m"
-    [[ $eval -eq 1 ]] && eval ${cmd}
-fi
-
-data_dir=${data_dir}/${lang}
-
-if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    echo "stage 1: ST Network Training"
-    [[ ! -d ${data_dir} ]] && echo "The data dir ${data_dir} is not existing!" && exit 1;
-
-    if [[ -z ${device} || ${#device[@]} -eq 0 ]]; then
-		if [[ ${gpu_num} -eq 0 ]]; then
-			device=""
-		else
-        	source ./local/utils.sh
-        	device=$(get_devices $gpu_num 0)
-		fi
-    fi
-
-    echo -e "dev=${device} data=${data_dir} model=${model_dir}"
-
-    if [[ ! -d ${model_dir} ]]; then
-        mkdir -p ${model_dir}
-    else
-        echo "${model_dir} exists."
-    fi
-
-    cp ${BASH_SOURCE[0]} ${model_dir}
-    cp ${PWD}/train.sh ${model_dir}
-
-    config_list="${train_config//,/ }"
-    idx=0
-    for config in ${config_list[@]}
-    do
-        config_path=$pwd_dir/conf/${config}.yaml
-        if [[ ! -f ${config_path} ]]; then
-            echo "No config file ${config_path}"
-            exit
-        fi
-        cp ${config_path} ${model_dir}
-
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
-        --train-config${idx} ${config_path}"
-        fi
-        idx=$((idx + 1))
-    done
-
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
-        ${data_dir}
-        --config-yaml ${data_config}
-        --task ${task}
-        --max-tokens ${max_tokens}
-        --skip-invalid-size-inputs-valid-test
-        --update-freq ${update_freq}
-        --log-interval 100
-        --save-dir ${model_dir}
-        --tensorboard-logdir ${model_dir}"
-
-	if [[ -n ${extra_parameter} ]]; then
-        cmd="${cmd}
-        ${extra_parameter}"
-    fi
-	if [[ ${gpu_num} -gt 0 ]]; then
-		cmd="${cmd}
-        --distributed-world-size $gpu_num
-        --ddp-backend no_c10d"
-	fi
-    if [[ $fp16 -eq 1 ]]; then
-        cmd="${cmd}
-        --fp16"
-    fi
-    if [[ $step_valid -eq 1 ]]; then
-        validate_interval=1
-        save_interval=1
-        keep_last_epochs=10
-        no_epoch_checkpoints=0
-        save_interval_updates=500
-        keep_interval_updates=10
-    else
-        validate_interval=1
-        keep_last_epochs=10
-    fi
-    if [[ $bleu_valid -eq 1 ]]; then
-        cmd="$cmd
-        --eval-bleu
-        --eval-bleu-args '{\"beam\": 1}'
-        --eval-tokenized-bleu
-        --eval-bleu-remove-bpe
-        --best-checkpoint-metric bleu
-        --maximize-best-checkpoint-metric"
-    fi
-    if [[ -n $no_epoch_checkpoints && $no_epoch_checkpoints -eq 1 ]]; then
-        cmd="$cmd
-        --no-epoch-checkpoints"
-    fi
-    if [[ -n $validate_interval ]]; then
-        cmd="${cmd}
-        --validate-interval $validate_interval "
-    fi
-    if [[ -n $save_interval ]]; then
-        cmd="${cmd}
-        --save-interval $save_interval "
-    fi
-    if [[ -n $keep_last_epochs ]]; then
-        cmd="${cmd}
-        --keep-last-epochs $keep_last_epochs "
-    fi
-    if [[ -n $save_interval_updates ]]; then
-        cmd="${cmd}
-        --save-interval-updates $save_interval_updates"
-        if [[ -n $keep_interval_updates ]]; then
-        cmd="${cmd}
-        --keep-interval-updates $keep_interval_updates"
-        fi
-    fi
-
-    echo -e "\033[34mRun command: \n${cmd} \033[0m"
-
-    # save info
-    log=./history.log
-    echo "${time} | ${device} | ${data_dir} | ${exp_name} | ${model_dir} " >> $log
-    tail -n 50 ${log} > tmp.log
-    mv tmp.log $log
-    export CUDA_VISIBLE_DEVICES=${device}
-
-    cmd="nohup ${cmd} >> ${model_dir}/train.log 2>&1 &"
-    if [[ $eval -eq 1 ]]; then
-		eval $cmd
-		sleep 2s
-		tail -n "$(wc -l ${model_dir}/train.log | awk '{print $1+1}')" -f ${model_dir}/train.log
-	fi
-fi
-wait
-
-if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    echo "stage 2: ST Decoding"
-    if [[ ${n_average} -ne 1 ]]; then
-        # Average models
-		dec_model=avg_${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 $cmd
-	else
-		dec_model=${dec_model}
-	fi
-
-    if [[ -z ${device} || ${#device[@]} -eq 0 ]]; then
-		if [[ ${gpu_num} -eq 0 ]]; then
-			device=""
-		else
-        	source ./local/utils.sh
-        	device=$(get_devices $gpu_num 0)
-		fi
-    fi
-    export CUDA_VISIBLE_DEVICES=${device}
-
-	result_file=${model_dir}/decode_result
-	[[ -f ${result_file} ]] && rm ${result_file}
-
-    test_subset=${test_subset//,/ }
-	for subset in "${test_subset[@]}"; do
-        subset=${subset}_st
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
-        ${data_dir}
-        --config-yaml ${data_config}
-        --gen-subset ${subset}
-        --task speech_to_text
-        --path ${model_dir}/${dec_model}
-        --results-path ${model_dir}
-        --max-tokens ${max_tokens}
-        --beam ${beam_size}
-        --lenpen ${len_penalty}
-        --scoring sacrebleu"
-    	echo -e "\033[34mRun command: \n${cmd} \033[0m"
-
-        if [[ $eval -eq 1 ]]; then
-    	    eval $cmd
-    	    tail -n 1 ${model_dir}/generate-${subset}.txt >> ${result_file}
-        fi
-	done
-    cat ${result_file}
-fi

egs/libri_trans/st/train.sh

-#! /bin/bash
-
-# training the model
-
-gpu_num=8
-update_freq=1
-max_tokens=40000
-
-extra_tag=
-extra_parameter=
-#extra_tag="${extra_tag}"
-#extra_parameter="${extra_parameter} "
-
-exp_tag=
-
-#config_list=(base)
-config_list=(ctc)
-#config_list=(sate_ctc)
-#config_list=(ctc conformer rpr)
-#config_list=(base sate)
-
-#config_list=(pds_base)
-#config_list=(pds_base conformer)
-
-# exp full name
-exp_name=
-
-train_config=$(echo ${config_list[*]} | sed 's/ /,/g')
-
-cmd="./run.sh
-    --stage 1
-    --stop_stage 1
-    --gpu_num ${gpu_num}
-    --update_freq ${update_freq}
-    --train_config ${train_config}
-    --max_tokens ${max_tokens}
-    "
-
-if [[ -n ${exp_name} ]]; then
-    cmd="$cmd --exp_name ${exp_name}"
-fi
-if [[ -n ${exp_tag} ]]; then
-    cmd="$cmd --exp_tag ${exp_tag}"
-fi
-if [[ -n ${extra_tag} ]]; then
-    cmd="$cmd --extra_tag ${extra_tag}"
-fi
-if [[ -n ${extra_parameter} ]]; then
-    cmd="$cmd --extra_parameter \"${extra_parameter}\""
-fi
-
-echo ${cmd}
-eval ${cmd}

egs/librispeech/asr/conf/base.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
 arch: s2t_transformer_s
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -39,4 +23,4 @@ encoder-attention-heads: 4
 
 
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 4
\ No newline at end of file

egs/librispeech/asr/conf/basis.yaml

+train-subset: train-clean-100,train-clean-360,train-other-500
+valid-subset: dev-clean
+
+max-epoch: 100
+max-update: 300000
+patience: 20
+best-checkpoint-metric: loss
+maximize-best-checkpoint-metric: False
+
+no-epoch-checkpoints: True
+#keep-last-epochs: 10
+keep-best-checkpoints: 10
+
+num-workers: 8
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+skip-invalid-size-inputs-valid-test: True
\ No newline at end of file

egs/librispeech/asr/conf/big.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: 20
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
 arch: s2t_transformer_m
 arch: s2t_transformer_m
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/conf/debug.yaml

@@ -2,19 +2,19 @@
 #arch: s2t_transformer_s
 #arch: s2t_transformer_s
 arch: s2t_sate
 arch: s2t_sate
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 2_2_6_2
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 cl-dropout: True
 cl-dropout: True
 cl-dropout-epoch: 50
 cl-dropout-epoch: 50

egs/librispeech/asr/conf/debug2.yaml

-train-subset: train-clean-100
-valid-subset: dev-clean
-
-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:
-#load-pretrained-acoustic-encoder-from:
-#load-pretrained-text-encoder-from:
-#load-pretrained-decoder-from:
-#load-pretrained-acoustic-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1007_st_ctc_baseline/avg_10_checkpoint.pt
-#load-pretrained-acoustic-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1111_st_ctc_conformer_lr0.001/avg_10_checkpoint.pt
-
-#load-pretrained-acoustic-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1007_st_pyramid4_all256_3333_sr8_ctc/avg_10_checkpoint.pt
-#load-pretrained-acoustic-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1114_st_pyramid4_all256_ctc_fix/avg_10_checkpoint.pt
-
-#load-pretrained-acoustic-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1015_st_pyramid4_all256_conformer_baseline/avg_10_checkpoint.pt
-#load-pretrained-acoustic-encoder-from: /home/xuchen/st/checkpoints/mustc/asr/1111_st_pyramid4_all256_conformer_ctc/avg_10_checkpoint.pt
-
-arch: s2t_sate
-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
-
-
-dropout: 0.1
-activation-fn: relu
-encoder-ffn-embed-dim: 2048
-encoder-layers: 12
-text-encoder-layers: 6
-decoder-layers: 6
-encoder-attention-heads: 4
-
-#macaron-style: True
-#use-cnn-module: True
-#cnn-module-kernel: 31
-
-#acoustic-encoder: transformer
-#acoustic-encoder: conformer
-acoustic-encoder: pyramid
-adapter: league
-#adapter: none
-#adapter: context
-
-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
\ No newline at end of file

egs/librispeech/asr/conf/pds_base.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
-
-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:
-
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
 clip-norm: 10.0
 clip-norm: 10.0
@@ -37,4 +20,4 @@ encoder-attention-heads: 4
 
 
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 4
\ No newline at end of file

egs/librispeech/asr/conf/pds_base_16.yaml

 arch: pdss2t_transformer_s_16
 arch: pdss2t_transformer_s_16
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 2_2_6_2
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -52,4 +36,4 @@ encoder-attention-heads: 4
 
 
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 4
\ No newline at end of file

egs/librispeech/asr/conf/pds_base_32.yaml

 arch: pdss2t_transformer_s_32
 arch: pdss2t_transformer_s_32
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 2_2_3_3_2
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256_256
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 8_8_8_8_8
-pyramid-attn-heads: 4_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: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
+pds-stages: 5
+#pds-dropout: 0
+pds-layers: 2_2_3_3_2
+pds-ratios: 2_2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1_1
+pds-kernel-sizes: 5_5_5_5_5
+pds-ffn-ratios: 8_8_8_8_8
+pds-attn-heads: 4_4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -52,4 +36,4 @@ encoder-attention-heads: 4
 
 
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 4
\ No newline at end of file

egs/librispeech/asr/conf/pds_base_8.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 3_3_3_3
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -52,4 +36,4 @@ encoder-attention-heads: 4
 
 
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 4
\ No newline at end of file

egs/librispeech/asr/conf/pds_big.yaml

@@ -2,22 +2,6 @@ arch: pdss2t_transformer_m_8
 #arch: pdss2t_transformer_m_16
 #arch: pdss2t_transformer_m_16
 #arch: pdss2t_transformer_m_32
 #arch: pdss2t_transformer_m_32
 
 
-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:
-
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
 clip-norm: 10.0
 clip-norm: 10.0

egs/librispeech/asr/conf/pds_big_16.yaml

 arch: pdss2t_transformer_m_16
 arch: pdss2t_transformer_m_16
 
 
 encoder-embed-dim: 512
 encoder-embed-dim: 512
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 512_512_512_512
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5
-pyramid-ffn-ratios: 4_4_4_4
-pyramid-attn-heads: 8_8_8_8
-
-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:
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 2_2_6_2
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 512_512_512_512
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 4_4_4_4
+pds-attn-heads: 8_8_8_8
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/conf/pds_big_32.yaml

 arch: pdss2t_transformer_m_32
 arch: pdss2t_transformer_m_32
 
 
 encoder-embed-dim: 512
 encoder-embed-dim: 512
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 2_2_3_3_2
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 512_512_512_512_512
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 4_4_4_4_4
-pyramid-attn-heads: 8_8_8_8_8
-
-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:
+pds-stages: 5
+#pds-dropout: 0
+pds-layers: 2_2_3_3_2
+pds-ratios: 2_2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 512_512_512_512_512
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1_1
+pds-kernel-sizes: 5_5_5_5_5
+pds-ffn-ratios: 4_4_4_4_4
+pds-attn-heads: 8_8_8_8_8
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/conf/pds_big_8.yaml

 arch: pdss2t_transformer_m_8
 arch: pdss2t_transformer_m_8
 
 
 encoder-embed-dim: 512
 encoder-embed-dim: 512
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 512_512_512_512
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5
-pyramid-ffn-ratios: 4_4_4_4
-pyramid-attn-heads: 8_8_8_8
-
-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:
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 3_3_3_3
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 512_512_512_512
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 4_4_4_4
+pds-attn-heads: 8_8_8_8
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/conf/pds_deep.yaml

@@ -2,22 +2,6 @@ arch: pdss2t_transformer_sd_8
 #arch: pdss2t_transformer_sd_16
 #arch: pdss2t_transformer_sd_16
 #arch: pdss2t_transformer_sd_32
 #arch: pdss2t_transformer_sd_32
 
 
-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:
-
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
 clip-norm: 10.0
 clip-norm: 10.0

egs/librispeech/asr/conf/pds_deep_16.yaml

 arch: pdss2t_transformer_sd_16
 arch: pdss2t_transformer_sd_16
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 5_5_12_8
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 5_5_12_8
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/conf/pds_deep_32.yaml

 arch: pdss2t_transformer_sd_32
 arch: pdss2t_transformer_sd_32
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 5_5_7_7_6
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256_256
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 8_8_8_8_8
-pyramid-attn-heads: 4_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: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
+pds-stages: 5
+#pds-dropout: 0
+pds-layers: 5_5_7_7_6
+pds-ratios: 2_2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1_1
+pds-kernel-sizes: 5_5_5_5_5
+pds-ffn-ratios: 8_8_8_8_8
+pds-attn-heads: 4_4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/conf/pds_deep_8.yaml

 arch: pdss2t_transformer_sd_8
 arch: pdss2t_transformer_sd_8
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 7_7_7_9
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 7_7_7_9
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/librispeech/asr/run.sh

@@ -24,7 +24,8 @@ device=()
 gpu_num=8
 gpu_num=8
 update_freq=1
 update_freq=1
 
 
-root_dir=~/st/Fairseq-S2T
+root_dir=~/st
+code_dir=${root_dir}/Fairseq-S2T
 pwd_dir=$PWD
 pwd_dir=$PWD
 
 
 # dataset
 # dataset
@@ -42,8 +43,8 @@ specific_prefix=valid
 specific_dir=/home/xuchen/st/data/mustc/st_lcrm/en-de
 specific_dir=/home/xuchen/st/data/mustc/st_lcrm/en-de
 asr_vocab_prefix=spm_unigram10000_st_share
 asr_vocab_prefix=spm_unigram10000_st_share
 
 
-org_data_dir=~/st/data/${dataset}
-data_dir=~/st/data/${dataset}
+org_data_dir=${root_dir}/data/${dataset}
+data_dir=${root_dir}/data/${dataset}
 test_subset=dev-clean,dev-other,test-clean,test-other
 test_subset=dev-clean,dev-other,test-clean,test-other
 
 
 # exp
 # exp
@@ -81,13 +82,12 @@ fi
 
 
 if [[ -z ${exp_name} ]]; then
 if [[ -z ${exp_name} ]]; then
     config_string=${train_config//,/_}
     config_string=${train_config//,/_}
-#    exp_name=${exp_prefix}_$(basename ${train_config%.*})_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     if [[ -n ${extra_tag} ]]; then
     if [[ -n ${extra_tag} ]]; then
         exp_name=${exp_name}_${extra_tag}
         exp_name=${exp_name}_${extra_tag}
     fi
     fi
 fi
 fi
-model_dir=$root_dir/../checkpoints/$dataset/asr/${exp_name}
+model_dir=$code_dir/../checkpoints/$dataset/asr/${exp_name}
 
 
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
     echo "stage -1: Data Download"
     echo "stage -1: Data Download"
@@ -103,7 +103,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         mkdir -p ${data_dir}
         mkdir -p ${data_dir}
     fi
     fi
 
 
-    cmd="python ${root_dir}/examples/speech_to_text/prep_librispeech_data.py
+    cmd="python ${code_dir}/examples/speech_to_text/prep_librispeech_data.py
         --data-root ${org_data_dir}
         --data-root ${org_data_dir}
         --output-root ${data_dir}
         --output-root ${data_dir}
         --vocab-type ${vocab_type}
         --vocab-type ${vocab_type}
@@ -146,28 +146,26 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
 
 
+    extra_parameter="${extra_parameter}
+        --train-config ${pwd_dir}/conf/basis.yaml"
+    cp ${pwd_dir}/conf/basis.yaml ${model_dir}
     config_list="${train_config//,/ }"
     config_list="${train_config//,/ }"
-    idx=0
+    idx=1
     for config in ${config_list[@]}
     for config in ${config_list[@]}
     do
     do
-        config_path=$pwd_dir/conf/${config}.yaml
+        config_path=${pwd_dir}/conf/${config}.yaml
         if [[ ! -f ${config_path} ]]; then
         if [[ ! -f ${config_path} ]]; then
             echo "No config file ${config_path}"
             echo "No config file ${config_path}"
             exit
             exit
         fi
         fi
         cp ${config_path} ${model_dir}
         cp ${config_path} ${model_dir}
 
 
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
+        extra_parameter="${extra_parameter}
         --train-config${idx} ${config_path}"
         --train-config${idx} ${config_path}"
-        fi
         idx=$((idx + 1))
         idx=$((idx + 1))
     done
     done
 
 
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
+    cmd="python3 -u ${code_dir}/fairseq_cli/train.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --task ${task}
         --task ${task}
@@ -252,9 +250,9 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 		dec_model=avg_${n_average}_checkpoint.pt
 		dec_model=avg_${n_average}_checkpoint.pt
 
 
         if [[ ! -f ${model_dir}/${dec_model} ]]; then
         if [[ ! -f ${model_dir}/${dec_model} ]]; then
-            cmd="python ${root_dir}/scripts/average_checkpoints.py
+            cmd="python ${code_dir}/scripts/average_checkpoints.py
             --inputs ${model_dir}
             --inputs ${model_dir}
-            --num-epoch-checkpoints ${n_average}
+            --num-best-checkpoints ${n_average}
             --output ${model_dir}/${dec_model}"
             --output ${model_dir}/${dec_model}"
             echo -e "\033[34mRun command: \n${cmd} \033[0m"
             echo -e "\033[34mRun command: \n${cmd} \033[0m"
             [[ $eval -eq 1 ]] && eval $cmd
             [[ $eval -eq 1 ]] && eval $cmd
@@ -279,7 +277,7 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     test_subset=(${test_subset//,/ })
     test_subset=(${test_subset//,/ })
 	for subset in ${test_subset[@]}; do
 	for subset in ${test_subset[@]}; do
         subset=${subset}
         subset=${subset}
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
+  		cmd="python ${code_dir}/fairseq_cli/generate.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --gen-subset ${subset}
         --gen-subset ${subset}

egs/mustc/asr/conf/base.yaml

-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
-
 arch: s2t_transformer_s
 arch: s2t_transformer_s
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/asr/conf/basis.yaml

+train-subset: train
+valid-subset: dev
+
+max-epoch: 100
+max-update: 100000
+patience: 20
+best_checkpoint_metric: loss
+maximize_best_checkpoint_metric: False
+
+no-epoch-checkpoints: True
+#keep-last-epochs: 10
+keep-best-checkpoints: 10
+
+num-workers: 8
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+skip-invalid-size-inputs-valid-test: True
\ No newline at end of file

egs/mustc/asr/conf/ctc.yaml

 ctc-weight: 0.3
 ctc-weight: 0.3
+post-process: sentencepiece

egs/mustc/asr/conf/pds_base.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
+pds-fusion: True
+ctc-layer: 12
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/asr/conf/pds_base_16.yaml

 arch: pdss2t_transformer_s_16
 arch: pdss2t_transformer_s_16
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
+pds-stages: 4
+ctc-layer: 12
+pds-layers: 2_2_6_2
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/asr/conf/pds_base_32.yaml

 arch: pdss2t_transformer_s_32
 arch: pdss2t_transformer_s_32
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 2_2_3_3_2
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256_256
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 8_8_8_8_8
-pyramid-attn-heads: 4_4_4_4_4
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
+pds-stages: 5
+ctc-layer: 12
+pds-layers: 2_2_3_3_2
+pds-ratios: 2_2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1_1
+pds-kernel-sizes: 5_5_5_5_5
+pds-ffn-ratios: 8_8_8_8_8
+pds-attn-heads: 4_4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/asr/conf/pds_base_8.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
-
-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#load-pretrained-decoder-from:
+pds-stages: 4
+ctc-layer: 12
+pds-layers: 3_3_3_3
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/asr/run.sh

@@ -24,7 +24,8 @@ stop_stage=0
 gpu_num=8
 gpu_num=8
 update_freq=1
 update_freq=1
 
 
-root_dir=~/st/Fairseq-S2T
+root_dir=~/st
+code_dir=${root_dir}/Fairseq-S2T
 pwd_dir=$PWD
 pwd_dir=$PWD
 
 
 # dataset
 # dataset
@@ -41,13 +42,16 @@ lcrm=0
 tokenizer=0
 tokenizer=0
 use_raw_audio=0
 use_raw_audio=0
 
 
-use_specific_dict=0
+use_specific_dict=1
 specific_prefix=st
 specific_prefix=st
-specific_dir=/home/xuchen/st/data/mustc/st/en-de
+specific_dir=${root_dir}/data/mustc/st
 asr_vocab_prefix=spm_unigram10000_st_share
 asr_vocab_prefix=spm_unigram10000_st_share
 
 
-org_data_dir=~/st/data/${dataset}
-data_dir=~/st/data/${dataset}/asr
+org_data_dir=${root_dir}/data/${dataset}
+data_dir=${root_dir}/data/${dataset}/asr
+train_split=train
+valid_split=dev
+test_split=tst-COMMON
 test_subset=tst-COMMON
 test_subset=tst-COMMON
 
 
 # exp
 # exp
@@ -59,7 +63,7 @@ exp_name=
 
 
 # config
 # config
 train_config=ctc
 train_config=ctc
-data_config=config_asr.yaml
+data_config=config.yaml
 
 
 # training setting
 # training setting
 fp16=1
 fp16=1
@@ -97,13 +101,12 @@ fi
 
 
 if [[ -z ${exp_name} ]]; then
 if [[ -z ${exp_name} ]]; then
     config_string=${train_config//,/_}
     config_string=${train_config//,/_}
-#    exp_name=${exp_prefix}_$(basename ${train_config%.*})_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     if [[ -n ${extra_tag} ]]; then
     if [[ -n ${extra_tag} ]]; then
         exp_name=${exp_name}_${extra_tag}
         exp_name=${exp_name}_${extra_tag}
     fi
     fi
 fi
 fi
-model_dir=$root_dir/../checkpoints/$dataset/asr/${exp_name}
+model_dir=${code_dir}/../checkpoints/${dataset}/asr/${exp_name}
 
 
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
     echo "stage -1: Data Download"
     echo "stage -1: Data Download"
@@ -114,11 +117,23 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     ### Task dependent. You have to make data the following preparation part by yourself.
     ### Task dependent. You have to make data the following preparation part by yourself.
     ### But you can utilize Kaldi recipes in most cases
     ### But you can utilize Kaldi recipes in most cases
     echo "stage 0: ASR Data Preparation"
     echo "stage 0: ASR Data Preparation"
+    if [[ ! -e ${data_dir} ]]; then
+        mkdir -p ${data_dir}
+    fi
+    feature_zip=fbank80.zip
+    if [[ ${speed_perturb} -eq 1 ]]; then
+        feature_zip=fbank80_sp.zip
+    fi
+    if [[ ! -f ${data_dir}/${feature_zip} && -f ${data_dir}/../feature_zip ]]; then
+        ln -s ${data_dir}/../feature_zip ${data_dir}
+    fi
 
 
-    cmd="python ${root_dir}/examples/speech_to_text/prep_mustc_data.py
+    cmd="python ${code_dir}/examples/speech_to_text/prep_audio_data.py
         --data-root ${org_data_dir}
         --data-root ${org_data_dir}
         --output-root ${data_dir}
         --output-root ${data_dir}
         --task asr
         --task asr
+        --src-lang ${src_lang}
+        --splits ${valid_split},${test_split},${train_split}
         --vocab-type ${vocab_type}
         --vocab-type ${vocab_type}
         --vocab-size ${vocab_size}"
         --vocab-size ${vocab_size}"
 
 
@@ -127,7 +142,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         --raw"
         --raw"
     fi
     fi
     if [[ ${use_specific_dict} -eq 1 ]]; then
     if [[ ${use_specific_dict} -eq 1 ]]; then
-        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}/${lang}
+        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}
         cmd="$cmd
         cmd="$cmd
         --asr-prefix ${asr_vocab_prefix}"
         --asr-prefix ${asr_vocab_prefix}"
     fi
     fi
@@ -147,6 +162,11 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
 
 
     echo -e "\033[34mRun command: \n${cmd} \033[0m"
     echo -e "\033[34mRun command: \n${cmd} \033[0m"
     [[ $eval -eq 1 ]] && eval ${cmd}
     [[ $eval -eq 1 ]] && eval ${cmd}
+
+    if [[ -f ${data_dir}/${feature_zip} && ! -f ${data_dir}/../${feature_zip} ]]; then
+        mv ${data_dir}/${feature_zip} ${data_dir}/..
+        ln -s ${data_dir}/../${feature_zip} ${data_dir}
+    fi
 fi
 fi
 
 
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
@@ -173,28 +193,26 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
 
 
+    extra_parameter="${extra_parameter}
+        --train-config ${pwd_dir}/conf/basis.yaml"
+    cp ${pwd_dir}/conf/basis.yaml ${model_dir}
     config_list="${train_config//,/ }"
     config_list="${train_config//,/ }"
-    idx=0
+    idx=1
     for config in ${config_list[@]}
     for config in ${config_list[@]}
     do
     do
-        config_path=$pwd_dir/conf/${config}.yaml
+        config_path=${pwd_dir}/conf/${config}.yaml
         if [[ ! -f ${config_path} ]]; then
         if [[ ! -f ${config_path} ]]; then
             echo "No config file ${config_path}"
             echo "No config file ${config_path}"
             exit
             exit
         fi
         fi
         cp ${config_path} ${model_dir}
         cp ${config_path} ${model_dir}
 
 
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
+        extra_parameter="${extra_parameter}
         --train-config${idx} ${config_path}"
         --train-config${idx} ${config_path}"
-        fi
         idx=$((idx + 1))
         idx=$((idx + 1))
     done
     done
 
 
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
+    cmd="python3 -u ${code_dir}/fairseq_cli/train.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --task ${task}
         --task ${task}
@@ -278,12 +296,14 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         # Average models
         # Average models
 		dec_model=avg_${n_average}_checkpoint.pt
 		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 ${code_dir}/scripts/average_checkpoints.py
+            --inputs ${model_dir}
+            --num-best-checkpoints ${n_average}
+            --output ${model_dir}/${dec_model}"
+            echo -e "\033[34mRun command: \n${cmd} \033[0m"
+            [[ $eval -eq 1 ]] && eval $cmd
+        fi
 	else
 	else
 		dec_model=${dec_model}
 		dec_model=${dec_model}
 	fi
 	fi
@@ -303,8 +323,8 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 
 
     test_subset=${test_subset//,/ }
     test_subset=${test_subset//,/ }
 	for subset in ${test_subset[@]}; do
 	for subset in ${test_subset[@]}; do
-        subset=${subset}_asr
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
+        subset=${subset}
+  		cmd="python ${code_dir}/fairseq_cli/generate.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --gen-subset ${subset}
         --gen-subset ${subset}

egs/mustc/mt/conf/base.yaml

-train-subset: train
-valid-subset: valid
-
-max-epoch: 50
-max-update: 100000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-skip-invalid-size-inputs-valid-test: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
 arch: transformer
 arch: transformer
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/mt/conf/base_s.yaml

-train-subset: train
-valid-subset: valid
-
-max-epoch: 50
-max-update: 100000
-
-num-workers: 8
-patience: 10
-no-progress-bar: True
-log-interval: 100
-seed: 1
-report-accuracy: True
-skip-invalid-size-inputs-valid-test: True
-
-#load-pretrained-encoder-from:
-#load-pretrained-decoder-from:
-
 arch: transformer
 arch: transformer
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam

egs/mustc/mt/conf/basis.yaml

+train-subset: train
+valid-subset: valid
+
+max-epoch: 50
+max-update: 100000
+patience: 20
+best_checkpoint_metric: loss
+maximize_best_checkpoint_metric: False
+
+no-epoch-checkpoints: True
+#keep-last-epochs: 10
+keep-best-checkpoints: 10
+
+num-workers: 8
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+skip-invalid-size-inputs-valid-test: True
\ No newline at end of file

egs/mustc/mt/decode.sh

@@ -10,6 +10,7 @@ if [ "$#" -eq 1 ]; then
     exp_name=$1
     exp_name=$1
 fi
 fi
 
 
+sacrebleu=1
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
 len_penalty=1.0
 len_penalty=1.0
@@ -21,6 +22,7 @@ cmd="./run.sh
     --stop_stage 2
     --stop_stage 2
     --gpu_num ${gpu_num}
     --gpu_num ${gpu_num}
     --exp_name ${exp_name}
     --exp_name ${exp_name}
+    --sacrebleu ${sacrebleu}
     --n_average ${n_average}
     --n_average ${n_average}
     --beam_size ${beam_size}
     --beam_size ${beam_size}
     --len_penalty ${len_penalty}
     --len_penalty ${len_penalty}

egs/mustc/mt/run.sh

@@ -13,7 +13,7 @@ set -o pipefail
 export PYTHONIOENCODING=UTF-8
 export PYTHONIOENCODING=UTF-8
 
 
 eval=1
 eval=1
-time=$(date "+%m%d_%H%M")
+time=$(date "+%m%d")
 
 
 stage=0
 stage=0
 stop_stage=0
 stop_stage=0
@@ -24,7 +24,8 @@ device=()
 gpu_num=8
 gpu_num=8
 update_freq=1
 update_freq=1
 
 
-root_dir=~/st/Fairseq-S2T
+root_dir=~/st
+code_dir=${root_dir}/Fairseq-S2T
 pwd_dir=$PWD
 pwd_dir=$PWD
 
 
 # dataset
 # dataset
@@ -42,12 +43,12 @@ tokenizer=0
 
 
 use_specific_dict=0
 use_specific_dict=0
 specific_prefix=st
 specific_prefix=st
-specific_dir=/home/xuchen/st/data/mustc/st/en-de/
+specific_dir=${root_dir}/data/mustc/st
 src_vocab_prefix=spm_unigram10000_st_share
 src_vocab_prefix=spm_unigram10000_st_share
 tgt_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/${lang}
+org_data_dir=${root_dir}/data/${dataset}
+data_dir=${root_dir}/data/${dataset}/mt
 train_subset=train
 train_subset=train
 valid_subset=dev
 valid_subset=dev
 trans_subset=tst-COMMON
 trans_subset=tst-COMMON
@@ -70,6 +71,7 @@ step_valid=0
 bleu_valid=0
 bleu_valid=0
 
 
 # decoding setting
 # decoding setting
+sacrebleu=1
 dec_model=checkpoint_best.pt
 dec_model=checkpoint_best.pt
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
@@ -106,7 +108,6 @@ fi
 # full path
 # full path
 if [[ -z ${exp_name} ]]; then
 if [[ -z ${exp_name} ]]; then
     config_string=${train_config//,/_}
     config_string=${train_config//,/_}
-#    exp_name=${exp_prefix}_$(basename ${train_config%.*})_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     if [[ -n ${extra_tag} ]]; then
     if [[ -n ${extra_tag} ]]; then
         exp_name=${exp_name}_${extra_tag}
         exp_name=${exp_name}_${extra_tag}
@@ -128,7 +129,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
 
 
     if [[ ! -f ${data_dir}/${src_vocab_prefix}.txt || ! -f ${data_dir}/${tgt_vocab_prefix}.txt ]]; then
     if [[ ! -f ${data_dir}/${src_vocab_prefix}.txt || ! -f ${data_dir}/${tgt_vocab_prefix}.txt ]]; then
         if [[ ${use_specific_dict} -eq 0 ]]; then
         if [[ ${use_specific_dict} -eq 0 ]]; then
-            cmd="python ${root_dir}/examples/speech_to_text/prep_mt_data.py
+            cmd="python ${code_dir}/examples/speech_to_text/prep_mt_data.py
                 --data-root ${org_data_dir}
                 --data-root ${org_data_dir}
                 --output-root ${data_dir}
                 --output-root ${data_dir}
                 --splits ${train_subset},${valid_subset},${trans_subset}
                 --splits ${train_subset},${valid_subset},${trans_subset}
@@ -151,9 +152,10 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     mkdir -p ${data_dir}/data
     mkdir -p ${data_dir}/data
     for split in ${train_subset} ${valid_subset} ${trans_subset}; do
     for split in ${train_subset} ${valid_subset} ${trans_subset}; do
     {
     {
-        cmd="cat ${org_data_dir}/${lang}/data/${split}/txt/${split}.${src_lang}"
+        txt_dir=${org_data_dir}/data/${split}/txt
+        cmd="cat ${txt_dir}/${split}.${src_lang}"
         if [[ ${lcrm} -eq 1 ]]; then
         if [[ ${lcrm} -eq 1 ]]; then
-            cmd="python local/lower_rm.py ${org_data_dir}/${lang}/data/${split}.${src_lang}"
+            cmd="python local/lower_rm.py ${org_data_dir}/data/${split}.${src_lang}"
         fi
         fi
         cmd="${cmd}
         cmd="${cmd}
         | spm_encode --model ${data_dir}/${src_vocab_prefix}.model
         | spm_encode --model ${data_dir}/${src_vocab_prefix}.model
@@ -166,7 +168,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         cmd="spm_encode
         cmd="spm_encode
         --model ${data_dir}/${tgt_vocab_prefix}.model
         --model ${data_dir}/${tgt_vocab_prefix}.model
         --output_format=piece
         --output_format=piece
-        < ${org_data_dir}/${lang}/data/${split}/txt/${split}.${tgt_lang}
+        < ${txt_dir}/${split}.${tgt_lang}
         > ${data_dir}/data/${split}.${tgt_lang}"
         > ${data_dir}/data/${split}.${tgt_lang}"
 
 
         echo -e "\033[34mRun command: \n${cmd} \033[0m"
         echo -e "\033[34mRun command: \n${cmd} \033[0m"
@@ -175,7 +177,7 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     done
     done
     wait
     wait
 
 
-    cmd="python ${root_dir}/fairseq_cli/preprocess.py
+    cmd="python ${code_dir}/fairseq_cli/preprocess.py
         --source-lang ${src_lang} --target-lang ${tgt_lang}
         --source-lang ${src_lang} --target-lang ${tgt_lang}
         --trainpref ${data_dir}/data/${train_subset}
         --trainpref ${data_dir}/data/${train_subset}
         --validpref ${data_dir}/data/${valid_subset}
         --validpref ${data_dir}/data/${valid_subset}
@@ -215,28 +217,26 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
 
 
+    extra_parameter="${extra_parameter}
+        --train-config ${pwd_dir}/conf/basis.yaml"
+    cp ${pwd_dir}/conf/basis.yaml ${model_dir}
     config_list="${train_config//,/ }"
     config_list="${train_config//,/ }"
-    idx=0
+    idx=1
     for config in ${config_list[@]}
     for config in ${config_list[@]}
     do
     do
-        config_path=$pwd_dir/conf/${config}.yaml
+        config_path=${pwd_dir}/conf/${config}.yaml
         if [[ ! -f ${config_path} ]]; then
         if [[ ! -f ${config_path} ]]; then
             echo "No config file ${config_path}"
             echo "No config file ${config_path}"
             exit
             exit
         fi
         fi
         cp ${config_path} ${model_dir}
         cp ${config_path} ${model_dir}
 
 
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
+        extra_parameter="${extra_parameter}
         --train-config${idx} ${config_path}"
         --train-config${idx} ${config_path}"
-        fi
         idx=$((idx + 1))
         idx=$((idx + 1))
     done
     done
 
 
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
+    cmd="python3 -u ${code_dir}/fairseq_cli/train.py
         ${data_dir}
         ${data_dir}
         --source-lang ${src_lang}
         --source-lang ${src_lang}
         --target-lang ${tgt_lang}
         --target-lang ${tgt_lang}
@@ -330,12 +330,14 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         # Average models
         # Average models
 		dec_model=avg_${n_average}_checkpoint.pt
 		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 ${code_dir}/scripts/average_checkpoints.py
+            --inputs ${model_dir}
+            --num-best-checkpoints ${n_average}
+            --output ${model_dir}/${dec_model}"
+            echo -e "\033[34mRun command: \n${cmd} \033[0m"
+            [[ $eval -eq 1 ]] && eval $cmd
+        fi
 	else
 	else
 		dec_model=${dec_model}
 		dec_model=${dec_model}
 	fi
 	fi
@@ -355,7 +357,7 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 
 
     test_subset=(${test_subset//,/ })
     test_subset=(${test_subset//,/ })
 	for subset in ${test_subset[@]}; do
 	for subset in ${test_subset[@]}; do
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
+  		cmd="python ${code_dir}/fairseq_cli/generate.py
         ${data_dir}
         ${data_dir}
         --source-lang ${src_lang}
         --source-lang ${src_lang}
         --target-lang ${tgt_lang}
         --target-lang ${tgt_lang}
@@ -366,14 +368,17 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         --max-tokens ${max_tokens}
         --max-tokens ${max_tokens}
         --beam ${beam_size}
         --beam ${beam_size}
         --lenpen ${len_penalty}
         --lenpen ${len_penalty}
-        --post-process sentencepiece
-        --scoring sacrebleu"
+        --post-process sentencepiece"
 
 
-        if [[ ${tokenizer} -eq 1 ]]; then
+        if [[ ${sacrebleu} -eq 1 ]]; then
+            cmd="${cmd}
+        --scoring sacrebleu"
+            if [[ ${tokenizer} -eq 1 ]]; then
             cmd="${cmd}
             cmd="${cmd}
         --tokenizer moses
         --tokenizer moses
         --moses-source-lang ${src_lang}
         --moses-source-lang ${src_lang}
         --moses-target-lang ${tgt_lang}"
         --moses-target-lang ${tgt_lang}"
+            fi
         fi
         fi
 
 
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"

egs/mustc/st/conf/base.yaml

-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:
-#load-pretrained-decoder-from:
-
 arch: s2t_transformer_s
 arch: s2t_transformer_s
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -42,3 +26,6 @@ decoder-ffn-embed-dim: 2048
 decoder-attention-heads: 4
 decoder-attention-heads: 4
 attention-dropout: 0.1
 attention-dropout: 0.1
 activation-dropout: 0.1
 activation-dropout: 0.1
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
\ No newline at end of file

egs/mustc/st/conf/basis.yaml

+train-subset: train
+valid-subset: dev
+
+max-epoch: 100
+max-update: 100000
+patience: 20
+best_checkpoint_metric: loss
+maximize_best_checkpoint_metric: False
+
+no-epoch-checkpoints: True
+#keep-last-epochs: 10
+keep-best-checkpoints: 10
+
+num-workers: 8
+no-progress-bar: True
+log-interval: 100
+seed: 1
+report-accuracy: True
+skip-invalid-size-inputs-valid-test: True
\ No newline at end of file

egs/mustc/st/conf/ctc.yaml

-ctc-weight: 0.3
+ctc-weight: 0.3
\ No newline at end of file
+post-process: sentencepiece
\ No newline at end of file

egs/mustc/st/conf/pds_base.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
-
-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:
-#load-pretrained-decoder-from:
+pds-fusion: True
+ctc-layer: 12
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -38,3 +24,6 @@ encoder-attention-heads: 4
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
 decoder-attention-heads: 4
 decoder-attention-heads: 4
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
\ No newline at end of file

egs/mustc/st/conf/pds_base_16.yaml

 arch: pdss2t_transformer_s_16
 arch: pdss2t_transformer_s_16
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 2_2_6_2
-pyramid-ratios: 2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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:
-#load-pretrained-decoder-from:
+pds-stages: 4
+ctc-layer: 12
+pds-layers: 2_2_6_2
+pds-ratios: 2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -53,3 +37,6 @@ encoder-attention-heads: 4
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
 decoder-attention-heads: 4
 decoder-attention-heads: 4
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
\ No newline at end of file

egs/mustc/st/conf/pds_base_32.yaml

 arch: pdss2t_transformer_s_32
 arch: pdss2t_transformer_s_32
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 5
-#pyramid-dropout: 0
-pyramid-layers: 2_2_3_3_2
-pyramid-ratios: 2_2_2_2_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256_256
-pyramid-ds-method: conv
-pyramid-embed-norm: True
-pyramid-position-embed: 1_1_1_1_1
-pyramid-kernel-sizes: 5_5_5_5_5
-pyramid-ffn-ratios: 8_8_8_8_8
-pyramid-attn-heads: 4_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:
-#load-pretrained-decoder-from:
+pds-stages: 5
+ctc-layer: 12
+pds-layers: 2_2_3_3_2
+pds-ratios: 2_2_2_2_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1_1
+pds-kernel-sizes: 5_5_5_5_5
+pds-ffn-ratios: 8_8_8_8_8
+pds-attn-heads: 4_4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -53,3 +37,6 @@ encoder-attention-heads: 4
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
 decoder-attention-heads: 4
 decoder-attention-heads: 4
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
\ No newline at end of file

egs/mustc/st/conf/pds_base_8.yaml

 arch: pdss2t_transformer_s_8
 arch: pdss2t_transformer_s_8
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-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:
-#load-pretrained-decoder-from:
+pds-stages: 4
+ctc-layer: 12
+pds-layers: 3_3_3_3
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -53,3 +37,6 @@ encoder-attention-heads: 4
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
 decoder-attention-heads: 4
 decoder-attention-heads: 4
+
+#load-pretrained-encoder-from:
+#load-pretrained-decoder-from:
\ No newline at end of file

egs/mustc/st/conf/sate_ctc.yaml

-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:
-#load-pretrained-acoustic-encoder-from:
-#load-pretrained-text-encoder-from:
-#load-pretrained-decoder-from:
-
 arch: s2t_sate
 arch: s2t_sate
 share-decoder-input-output-embed: True
 share-decoder-input-output-embed: True
 optimizer: adam
 optimizer: adam
@@ -43,6 +25,11 @@ text-encoder-layers: 6
 decoder-layers: 6
 decoder-layers: 6
 encoder-attention-heads: 4
 encoder-attention-heads: 4
 
 
+#load-pretrained-encoder-from:
+#load-pretrained-acoustic-encoder-from:
+#load-pretrained-text-encoder-from:
+#load-pretrained-decoder-from:
+
 #macaron-style: True
 #macaron-style: True
 #use-cnn-module: True
 #use-cnn-module: True
 #cnn-module-kernel: 31
 #cnn-module-kernel: 31
@@ -52,20 +39,20 @@ acoustic-encoder: transformer
 adapter: league
 adapter: league
 
 
 encoder-embed-dim: 256
 encoder-embed-dim: 256
-pyramid-stages: 4
-#pyramid-dropout: 0
-pyramid-layers: 3_3_3_3
-pyramid-ratios: 2_2_1_2
-pyramid-fusion: True
-pyramid-fusion-method: all_conv
-pyramid-embed-dims: 256_256_256_256
-pyramid-ds-method: 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-attn-heads: 4_4_4_4
+pds-stages: 4
+#pds-dropout: 0
+pds-layers: 3_3_3_3
+pds-ratios: 2_2_1_2
+pds-fusion: True
+pds-fusion-method: all_conv
+pds-embed-dims: 256_256_256_256
+pds-ds-method: conv
+pds-embed-norm: True
+pds-position-embed: 1_1_1_1
+pds-kernel-sizes: 5_5_5_5
+pds-ffn-ratios: 8_8_8_8
+pds-attn-heads: 4_4_4_4
 
 
 decoder-embed-dim: 256
 decoder-embed-dim: 256
 decoder-ffn-embed-dim: 2048
 decoder-ffn-embed-dim: 2048
-decoder-attention-heads: 4
+decoder-attention-heads: 4
\ No newline at end of file

egs/mustc/st/decode.sh

@@ -3,13 +3,14 @@
 gpu_num=1
 gpu_num=1
 
 
 data_dir=
 data_dir=
-test_subset=(tst-COMMON)
+test_subset=(dev tst-COMMON)
 
 
 exp_name=
 exp_name=
 if [ "$#" -eq 1 ]; then
 if [ "$#" -eq 1 ]; then
     exp_name=$1
     exp_name=$1
 fi
 fi
 
 
+sacrebleu=1
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
 len_penalty=1.0
 len_penalty=1.0
@@ -21,6 +22,7 @@ cmd="./run.sh
     --stop_stage 2
     --stop_stage 2
     --gpu_num ${gpu_num}
     --gpu_num ${gpu_num}
     --exp_name ${exp_name}
     --exp_name ${exp_name}
+    --sacrebleu ${sacrebleu}
     --n_average ${n_average}
     --n_average ${n_average}
     --beam_size ${beam_size}
     --beam_size ${beam_size}
     --len_penalty ${len_penalty}
     --len_penalty ${len_penalty}
@@ -31,7 +33,7 @@ cmd="./run.sh
 if [[ -n ${data_dir} ]]; then
 if [[ -n ${data_dir} ]]; then
     cmd="$cmd --data_dir ${data_dir}"
     cmd="$cmd --data_dir ${data_dir}"
 fi
 fi
-if [[ ${#test_subset[@]} -eq 0 ]]; then
+if [[ ${#test_subset[@]} -ne 0 ]]; then
     subsets=$(echo ${test_subset[*]} | sed 's/ /,/g')
     subsets=$(echo ${test_subset[*]} | sed 's/ /,/g')
     cmd="$cmd --test_subset ${subsets}"
     cmd="$cmd --test_subset ${subsets}"
 fi
 fi

egs/mustc/st/run.sh

@@ -24,7 +24,8 @@ stop_stage=0
 gpu_num=8
 gpu_num=8
 update_freq=1
 update_freq=1
 
 
-root_dir=~/st/Fairseq-S2T
+root_dir=~/st
+code_dir=${root_dir}/Fairseq-S2T
 pwd_dir=$PWD
 pwd_dir=$PWD
 
 
 # dataset
 # dataset
@@ -41,15 +42,19 @@ share_dict=1
 speed_perturb=0
 speed_perturb=0
 lcrm=0
 lcrm=0
 tokenizer=0
 tokenizer=0
+use_raw_audio=0
 
 
 use_specific_dict=0
 use_specific_dict=0
 specific_prefix=valid
 specific_prefix=valid
-specific_dir=/home/xuchen/st/data/mustc/st_lcrm/en-de
+specific_dir=${root_dir}/data/mustc/st
 asr_vocab_prefix=spm_unigram10000_st_share
 asr_vocab_prefix=spm_unigram10000_st_share
 st_vocab_prefix=spm_unigram10000_st_share
 st_vocab_prefix=spm_unigram10000_st_share
 
 
-org_data_dir=~/st/data/${dataset}
-data_dir=~/st/data/${dataset}/st
+org_data_dir=${root_dir}/data/${dataset}
+data_dir=${root_dir}/data/${dataset}/st
+train_split=train
+valid_split=dev
+test_split=tst-COMMON
 test_subset=tst-COMMON
 test_subset=tst-COMMON
 
 
 # exp
 # exp
@@ -60,7 +65,7 @@ exp_tag=baseline
 exp_name=
 exp_name=
 
 
 # config
 # config
-train_config=ctc
+train_config=base,ctc
 
 
 # training setting
 # training setting
 fp16=1
 fp16=1
@@ -69,15 +74,16 @@ step_valid=0
 bleu_valid=0
 bleu_valid=0
 
 
 # decoding setting
 # decoding setting
+sacrebleu=1
 dec_model=checkpoint_best.pt
 dec_model=checkpoint_best.pt
 n_average=10
 n_average=10
 beam_size=5
 beam_size=5
 len_penalty=1.0
 len_penalty=1.0
 
 
 if [[ ${share_dict} -eq 1 ]]; then
 if [[ ${share_dict} -eq 1 ]]; then
-	data_config=config_st_share.yaml
+	data_config=config_share.yaml
 else
 else
-	data_config=config_st.yaml
+	data_config=config.yaml
 fi
 fi
 if [[ ${speed_perturb} -eq 1 ]]; then
 if [[ ${speed_perturb} -eq 1 ]]; then
     data_dir=${data_dir}_sp
     data_dir=${data_dir}_sp
@@ -95,18 +101,21 @@ if [[ ${tokenizer} -eq 1 ]]; then
     data_dir=${data_dir}_tok
     data_dir=${data_dir}_tok
     exp_prefix=${exp_prefix}_tok
     exp_prefix=${exp_prefix}_tok
 fi
 fi
+if [[ ${use_raw_audio} -eq 1 ]]; then
+    data_dir=${data_dir}_raw
+    exp_prefix=${exp_prefix}_raw
+fi
 
 
 . ./local/parse_options.sh || exit 1;
 . ./local/parse_options.sh || exit 1;
 
 
 if [[ -z ${exp_name} ]]; then
 if [[ -z ${exp_name} ]]; then
     config_string=${train_config//,/_}
     config_string=${train_config//,/_}
-#    exp_name=${exp_prefix}_$(basename ${train_config%.*})_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     exp_name=${exp_prefix}_${config_string}_${exp_tag}
     if [[ -n ${extra_tag} ]]; then
     if [[ -n ${extra_tag} ]]; then
         exp_name=${exp_name}_${extra_tag}
         exp_name=${exp_name}_${extra_tag}
     fi
     fi
 fi
 fi
-model_dir=$root_dir/../checkpoints/$dataset/st/${exp_name}
+model_dir=$code_dir/../checkpoints/$dataset/st/${exp_name}
 
 
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
 if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
     echo "stage -1: Data Download"
     echo "stage -1: Data Download"
@@ -117,37 +126,49 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     ### Task dependent. You have to make data the following preparation part by yourself.
     ### Task dependent. You have to make data the following preparation part by yourself.
     ### But you can utilize Kaldi recipes in most cases
     ### But you can utilize Kaldi recipes in most cases
     echo "stage 0: ASR Data Preparation"
     echo "stage 0: ASR Data Preparation"
-    if [[ ! -e ${data_dir}/${lang} ]]; then
-        mkdir -p ${data_dir}/${lang}
+    if [[ ! -e ${data_dir} ]]; then
+        mkdir -p ${data_dir}
+    fi
+    if [[ ! -f ${data_dir}/fbank80.zip && -f ${data_dir}/../fbank80.zip ]]; then
+        ln -s ${data_dir}/../fbank80.zip ${data_dir}
+    fi
+    if [[ ! -f ${data_dir}/fbank80_sp.zip && -f ${data_dir}/../fbank80_sp.zip ]]; then
+        ln -s ${data_dir}/../fbank80_sp.zip ${data_dir}
     fi
     fi
 
 
-    cmd="python ${root_dir}/examples/speech_to_text/prep_mustc_data.py
+    # create ASR vocabulary if necessary
+    cmd="python ${code_dir}/examples/speech_to_text/prep_audio_data.py
         --data-root ${org_data_dir}
         --data-root ${org_data_dir}
-        --output-root ${data_dir}
+        --output-root ${data_dir}/asr4st
         --task asr
         --task asr
+        --raw
+        --src-lang ${src_lang}
+        --splits ${valid_split},${test_split},${train_split}
         --vocab-type ${vocab_type}
         --vocab-type ${vocab_type}
         --vocab-size ${asr_vocab_size}"
         --vocab-size ${asr_vocab_size}"
-    if [[ ${speed_perturb} -eq 1 ]]; then
-        cmd="$cmd
-        --speed-perturb"
-    fi
-    echo -e "\033[34mRun command: \n${cmd} \033[0m"
-    [[ $eval -eq 1 && ${share_dict} -ne 1 && ${use_specific_dict} -ne 1 ]] && eval $cmd
+    [[ $eval -eq 1 && ${share_dict} -ne 1 && ${use_specific_dict} -ne 1 ]] && (echo -e "\033[34mRun command: \n${cmd} \033[0m" && eval $cmd)
     asr_prefix=spm_${vocab_type}${asr_vocab_size}_asr
     asr_prefix=spm_${vocab_type}${asr_vocab_size}_asr
 
 
     echo "stage 0: ST Data Preparation"
     echo "stage 0: ST Data Preparation"
-    cmd="python ${root_dir}/examples/speech_to_text/prep_mustc_data.py
+    cmd="python ${code_dir}/examples/speech_to_text/prep_audio_data.py
         --data-root ${org_data_dir}
         --data-root ${org_data_dir}
         --output-root ${data_dir}
         --output-root ${data_dir}
         --task st
         --task st
         --add-src
         --add-src
+        --src-lang ${src_lang}
+        --tgt-lang ${tgt_lang}
+        --splits ${valid_split},${test_split},${train_split}
         --cmvn-type utterance
         --cmvn-type utterance
         --vocab-type ${vocab_type}
         --vocab-type ${vocab_type}
         --vocab-size ${vocab_size}"
         --vocab-size ${vocab_size}"
 
 
+    if [[ ${use_raw_audio} -eq 1 ]]; then
+        cmd="$cmd
+        --raw"
+    fi
     if [[ ${use_specific_dict} -eq 1 ]]; then
     if [[ ${use_specific_dict} -eq 1 ]]; then
-        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}/${lang}
-        cp -r ${specific_dir}/${st_vocab_prefix}.* ${data_dir}/${lang}
+        cp -r ${specific_dir}/${asr_vocab_prefix}.* ${data_dir}
+        cp -r ${specific_dir}/${st_vocab_prefix}.* ${data_dir}
         if [[ $share_dict -eq 1 ]]; then
         if [[ $share_dict -eq 1 ]]; then
             cmd="$cmd
             cmd="$cmd
         --share
         --share
@@ -182,9 +203,16 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
 
 
     echo -e "\033[34mRun command: \n${cmd} \033[0m"
     echo -e "\033[34mRun command: \n${cmd} \033[0m"
     [[ $eval -eq 1 ]] && eval ${cmd}
     [[ $eval -eq 1 ]] && eval ${cmd}
-fi
 
 
-data_dir=${data_dir}/${lang}
+    if [[ ! -f ${data_dir}/../fbank80.zip ]]; then
+        mv ${data_dir}/fbank80.zip ${data_dir}/..
+        ln -s ${data_dir}/../fbank80.zip ${data_dir}
+    fi
+    if [[ ! -f ${data_dir}/../fbank80_sp.zip ]]; then
+        mv ${data_dir}/fbank80_sp.zip ${data_dir}/..
+        ln -s ${data_dir}/../fbank80_sp.zip ${data_dir}
+    fi
+fi
 
 
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     echo "stage 1: ST Network Training"
     echo "stage 1: ST Network Training"
@@ -210,28 +238,26 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${BASH_SOURCE[0]} ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
     cp ${PWD}/train.sh ${model_dir}
 
 
+    extra_parameter="${extra_parameter}
+        --train-config ${pwd_dir}/conf/basis.yaml"
+    cp ${pwd_dir}/conf/basis.yaml ${model_dir}
     config_list="${train_config//,/ }"
     config_list="${train_config//,/ }"
-    idx=0
+    idx=1
     for config in ${config_list[@]}
     for config in ${config_list[@]}
     do
     do
-        config_path=$pwd_dir/conf/${config}.yaml
+        config_path=${pwd_dir}/conf/${config}.yaml
         if [[ ! -f ${config_path} ]]; then
         if [[ ! -f ${config_path} ]]; then
             echo "No config file ${config_path}"
             echo "No config file ${config_path}"
             exit
             exit
         fi
         fi
         cp ${config_path} ${model_dir}
         cp ${config_path} ${model_dir}
 
 
-        if [[ idx -eq 0 ]]; then
-            extra_parameter="${extra_parameter}
-        --train-config ${config_path}"
-        else
-            extra_parameter="${extra_parameter}
+        extra_parameter="${extra_parameter}
         --train-config${idx} ${config_path}"
         --train-config${idx} ${config_path}"
-        fi
         idx=$((idx + 1))
         idx=$((idx + 1))
     done
     done
 
 
-    cmd="python3 -u ${root_dir}/fairseq_cli/train.py
+    cmd="python3 -u ${code_dir}/fairseq_cli/train.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --task ${task}
         --task ${task}
@@ -324,12 +350,14 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         # Average models
         # Average models
 		dec_model=avg_${n_average}_checkpoint.pt
 		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 ${code_dir}/scripts/average_checkpoints.py
+            --inputs ${model_dir}
+            --num-best-checkpoints ${n_average}
+            --output ${model_dir}/${dec_model}"
+            echo -e "\033[34mRun command: \n${cmd} \033[0m"
+            [[ $eval -eq 1 ]] && eval $cmd
+        fi
 	else
 	else
 		dec_model=${dec_model}
 		dec_model=${dec_model}
 	fi
 	fi
@@ -348,9 +376,9 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 	[[ -f ${result_file} ]] && rm ${result_file}
 	[[ -f ${result_file} ]] && rm ${result_file}
 
 
     test_subset=${test_subset//,/ }
     test_subset=${test_subset//,/ }
-	for subset in "${test_subset[@]}"; do
-        subset=${subset}_st
-  		cmd="python ${root_dir}/fairseq_cli/generate.py
+	for subset in ${test_subset[@]}; do
+        subset=${subset}
+  		cmd="python ${code_dir}/fairseq_cli/generate.py
         ${data_dir}
         ${data_dir}
         --config-yaml ${data_config}
         --config-yaml ${data_config}
         --gen-subset ${subset}
         --gen-subset ${subset}
@@ -359,8 +387,19 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
         --results-path ${model_dir}
         --results-path ${model_dir}
         --max-tokens ${max_tokens}
         --max-tokens ${max_tokens}
         --beam ${beam_size}
         --beam ${beam_size}
-        --lenpen ${len_penalty}
+        --lenpen ${len_penalty}"
+
+        if [[ ${sacrebleu} -eq 1 ]]; then
+            cmd="${cmd}
         --scoring sacrebleu"
         --scoring sacrebleu"
+            if [[ ${tokenizer} -eq 1 ]]; then
+            cmd="${cmd}
+        --tokenizer moses
+        --moses-source-lang ${src_lang}
+        --moses-target-lang ${tgt_lang}"
+            fi
+        fi
+
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
     	echo -e "\033[34mRun command: \n${cmd} \033[0m"
 
 
         if [[ $eval -eq 1 ]]; then
         if [[ $eval -eq 1 ]]; then

egs/template/asr/binary.sh

-set -e
-
-eval=1
-
-lcrm=0
-tokenizer=0
-
-root_dir=~/st/Fairseq-S2T
-data_dir=~/st/data/test
-vocab_dir=~/st/data/mustc/st/en-de
-asr_vocab_prefix=spm_unigram10000_st_share
-
-src_lang=en
-tgt_lang=de
-subsets=(2019)
-
-cp -r ${vocab_dir}/${asr_vocab_prefix}.* ${data_dir}/${src_lang}-${tgt_lang}
-rm -rf ${data_dir}/${src_lang}-${tgt_lang}/fbank80.zip
-
-splits=$(echo ${subsets[*]} | sed 's/ /,/g')
-cmd="python ${root_dir}/examples/speech_to_text/prep_st_data.py
-    --data-root ${data_dir}
-    --output-root ${data_dir}
-    --splits ${splits}
-    --task asr
-    --src-lang ${src_lang}
-    --tgt-lang ${tgt_lang}
-    --add-src
-    --share
-    --asr-prefix ${asr_vocab_prefix}
-    --cmvn-type utterance"
-
-    if [[ ${lcrm} -eq 1 ]]; then
-        cmd="$cmd
-    --lowercase-src
-    --rm-punc-src"
-    fi
-    if [[ ${tokenizer} -eq 1 ]]; then
-        cmd="$cmd
-    --tokenizer"
-    fi
-
-echo -e "\033[34mRun command: \n${cmd} \033[0m"
-[[ $eval -eq 1 ]] && eval ${cmd}

egs/template/asr/conf/base.yaml

-train-subset: train_asr
-valid-subset: dev_asr
-
-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:
-#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)
-
-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/template/asr/conf/conformer.yaml

-macaron-style: True
-use-cnn-module: True
-cnn-module-kernel: 31

egs/template/asr/conf/ctc.yaml

-ctc-weight: 0.3

egs/template/asr/conf/dlcl.yaml

-use-enc-dlcl: True
-use-dec-dlcl: True

egs/template/asr/conf/local_attn.yaml

-encoder-attention-type: local
-hard-mask-window: 0
-gauss-mask-sigma: 3
-init-mask-weight: 0
\ No newline at end of file