update the shell scripts

egs/mustc/asr/conf/train_ctc.yaml

@@ -34,7 +34,7 @@ dropout: 0.1
 activation-fn: relu
 encoder-embed-dim: 256
 encoder-ffn-embed-dim: 2048
-encoder-layers: 12
+encoder-layers: 6
 decoder-layers: 6
 encoder-attention-heads: 4
 

egs/mustc/asr/run.sh

@@ -24,7 +24,7 @@ device=()
 gpu_num=8
 update_freq=1
 
-root_dir=~/st/fairseq
+root_dir=~/st/Fairseq-S2T
 pwd_dir=$PWD
 
 # dataset

egs/mustc/asr/train.sh

@@ -3,8 +3,8 @@
 # training the model
 
 gpu_num=8
-update_freq=1
-max_tokens=40000
+update_freq=2
+max_tokens=20000
 
 extra_tag=
 extra_parameter=
@@ -13,7 +13,7 @@ extra_parameter=
 #extra_parameter="${extra_parameter} "
 
 exp_tag=
-train_config=asr_train_ctc.yaml
+train_config=train_ctc.yaml
 
 cmd="./run.sh
     --stage 1

egs/mustc/mt/conf/train.yaml

@@ -21,7 +21,7 @@ clip-norm: 10.0
 lr-scheduler: inverse_sqrt
 warmup-init-lr: 1e-7
 warmup-updates: 8000
-lr: 1e-3
+lr: 5e-4
 adam_betas: (0.9,0.98)
 
 criterion: label_smoothed_cross_entropy

egs/mustc/mt/run.sh

@@ -24,7 +24,7 @@ device=()
 gpu_num=8
 update_freq=1
 
-root_dir=~/st/fairseq
+root_dir=~/st/Fairseq-S2T
 pwd_dir=$PWD
 
 # dataset
@@ -53,7 +53,7 @@ train_config=st_train_ctc.yaml
 
 # training setting
 fp16=1
-max_tokens=40000
+max_tokens=4096
 step_valid=0
 bleu_valid=0
 

egs/mustc/mt/train.sh

@@ -12,7 +12,7 @@ extra_parameter=
 #extra_tag="${extra_tag}"
 #extra_parameter="${extra_parameter} "
 
-exp_tag=
+exp_tag=baseline
 train_config=train.yaml
 
 cmd="./run.sh

egs/mustc/st/conf/train.yaml

@@ -33,7 +33,7 @@ dropout: 0.1
 activation-fn: relu
 encoder-embed-dim: 256
 encoder-ffn-embed-dim: 2048
-encoder-layers: 12
+encoder-layers: 6
 decoder-layers: 6
 encoder-attention-heads: 4
 

egs/mustc/st/conf/train_ctc.yaml

@@ -34,7 +34,7 @@ dropout: 0.1
 activation-fn: relu
 encoder-embed-dim: 256
 encoder-ffn-embed-dim: 2048
-encoder-layers: 12
+encoder-layers: 6
 decoder-layers: 6
 encoder-attention-heads: 4
 

egs/mustc/st/conf/train_ctc_conformer.yaml

 train-subset: train_st
 valid-subset: dev_st
 
-max-epoch: 100
+max-epoch: 50
 max-update: 100000
 
 num-workers: 8
@@ -34,7 +34,7 @@ dropout: 0.1
 activation-fn: relu
 encoder-embed-dim: 256
 encoder-ffn-embed-dim: 2048
-encoder-layers: 12
+encoder-layers: 6
 decoder-layers: 6
 encoder-attention-heads: 4
 

egs/mustc/st/run.sh

@@ -24,7 +24,7 @@ device=()
 gpu_num=8
 update_freq=1
 
-root_dir=~/st/fairseq
+root_dir=~/st/Fairseq-S2T
 pwd_dir=$PWD
 
 # dataset
@@ -101,8 +101,8 @@ 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} ]]; then
-        mkdir -p ${data_dir}
+    if [[ ! -e ${data_dir}/${lang} ]]; then
+        mkdir -p ${data_dir}/${lang}
     fi
 
     cmd="python ${root_dir}/examples/speech_to_text/prep_mustc_data.py
@@ -166,7 +166,7 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
         ${data_dir}
         --config-yaml ${data_config}
         --train-config ${train_config}
-        --task speech_to_text
+        --task ${task}
         --max-tokens ${max_tokens}
         --update-freq ${update_freq}
         --log-interval 100