update inference config

egs_modelscope/aishell/paraformer/paraformer_large_finetune.sh

@@ -7,7 +7,7 @@ CUDA_VISIBLE_DEVICES="0,1" # set gpus, e.g., CUDA_VISIBLE_DEVICES="0,1"
 gpu_num=2
 count=1
 gpu_inference=true # Whether to perform gpu decoding, set false for cpu decoding
-njob=4 # the number of jobs for each gpu
+njob=1 # the number of jobs for each gpu
 train_cmd=utils/run.pl
 infer_cmd=utils/run.pl
 
@@ -84,7 +84,7 @@ gpuid_list=$CUDA_VISIBLE_DEVICES  # set gpus for decoding, the same as training
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
 
 if ${gpu_inference}; then
-    inference_nj=$[${ngpu}*${njob}]
+    inference_nj=$njob
     _ngpu=1
 else
     inference_nj=$njob
@@ -237,10 +237,10 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
         fi
         ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \
             python -m funasr.bin.asr_inference_launch \
-                --batch_size 1 \
+                --batch_size 64 \
                 --ngpu "${_ngpu}" \
                 --njob ${njob} \
-                --gpuid_list ${gpuid_list} \
+                --gpuid_list ${gpuid_list:0:1} \
                 --data_path_and_name_and_type "${_data}/${scp},speech,${type}" \
                 --key_file "${_logdir}"/keys.JOB.scp \
                 --asr_train_config "${asr_exp}"/config.yaml \

egs_modelscope/aishell/paraformer/paraformer_large_infer.sh

@@ -10,9 +10,9 @@ exp_dir=
 model_name=speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch
 model_revision="v1.0.4"     # please do not modify the model revision
 inference_nj=32
-gpuid_list="0,1" # set gpus, e.g., gpuid_list="0,1"
+gpuid_list="0" # set gpus, e.g., gpuid_list="0,1"
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
-njob=4  # the number of jobs for each gpu
+njob=1  # the number of jobs for each gpu
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 
 if ${gpu_inference}; then

egs_modelscope/aishell2/paraformer/paraformer_large_finetune.sh

@@ -7,7 +7,7 @@ CUDA_VISIBLE_DEVICES="0,1" # set gpus, e.g., CUDA_VISIBLE_DEVICES="0,1"
 gpu_num=2
 count=1
 gpu_inference=true # Whether to perform gpu decoding, set false for cpu decoding
-njob=4 # the number of jobs for each gpu
+njob=1 # the number of jobs for each gpu
 train_cmd=utils/run.pl
 infer_cmd=utils/run.pl
 
@@ -85,7 +85,7 @@ gpuid_list=$CUDA_VISIBLE_DEVICES  # set gpus for decoding, the same as training
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
 
 if ${gpu_inference}; then
-    inference_nj=$[${ngpu}*${njob}]
+    inference_nj=$njob
     _ngpu=1
 else
     inference_nj=$njob
@@ -252,10 +252,10 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
         fi
         ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \
             python -m funasr.bin.asr_inference_launch \
-                --batch_size 1 \
+                --batch_size 64 \
                 --ngpu "${_ngpu}" \
                 --njob ${njob} \
-                --gpuid_list ${gpuid_list} \
+                --gpuid_list ${gpuid_list:0:1} \
                 --data_path_and_name_and_type "${_data}/${scp},speech,${type}" \
                 --key_file "${_logdir}"/keys.JOB.scp \
                 --asr_train_config "${asr_exp}"/config.yaml \

egs_modelscope/aishell2/paraformer/paraformer_large_infer.sh

@@ -10,9 +10,9 @@ exp_dir=
 model_name=speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch
 model_revision="v1.0.4"     # please do not modify the model revision
 inference_nj=32
-gpuid_list="0,1" # set gpus, e.g., gpuid_list="0,1"
+gpuid_list="0" # set gpus, e.g., gpuid_list="0,1"
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
-njob=4  # the number of jobs for each gpu
+njob=1  # the number of jobs for each gpu
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 
 if ${gpu_inference}; then

egs_modelscope/common/modelscope_common_finetune.sh

@@ -7,7 +7,7 @@ CUDA_VISIBLE_DEVICES="0,1" # set gpus, e.g., CUDA_VISIBLE_DEVICES="0,1"
 gpu_num=2
 count=1
 gpu_inference=true # Whether to perform gpu decoding, set false for cpu decoding
-njob=4 # the number of jobs for each gpu
+njob=1 # the number of jobs for each gpu
 train_cmd=utils/run.pl
 infer_cmd=utils/run.pl
 
@@ -84,7 +84,7 @@ gpuid_list=$CUDA_VISIBLE_DEVICES  # set gpus for decoding, the same as training
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
 
 if ${gpu_inference}; then
-    inference_nj=$[${ngpu}*${njob}]
+    inference_nj=$njob
     _ngpu=1
 else
     inference_nj=$njob
@@ -244,10 +244,10 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
         fi
         ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \
             python -m funasr.bin.asr_inference_launch \
-                --batch_size 1 \
+                --batch_size 64 \
                 --ngpu "${_ngpu}" \
                 --njob ${njob} \
-                --gpuid_list ${gpuid_list} \
+                --gpuid_list ${gpuid_list:0:1} \
                 --data_path_and_name_and_type "${_data}/${scp},speech,${type}" \
                 --key_file "${_logdir}"/keys.JOB.scp \
                 --asr_train_config "${asr_exp}"/config.yaml \

egs_modelscope/common/modelscope_common_infer.sh

@@ -8,9 +8,9 @@ model_name=speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch  #
 model_revision="v1.0.4"     # please do not modify the model revision
 data_dir=  # wav list, ${data_dir}/wav.scp
 exp_dir="exp"
-gpuid_list="0,1"
+gpuid_list="0"
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
-njob=4
+njob=1
 gpu_inference=true
 decode_cmd=utils/run.pl
 

egs_modelscope/common/modelscope_common_infer_after_finetune.sh

@@ -8,9 +8,9 @@ pretrained_model_name=speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404
 data_dir=  # wav list, ${data_dir}/wav.scp
 finetune_model_name=  # fine-tuning model name
 finetune_exp_dir=  # fine-tuning model experiment result path
-gpuid_list="0,1"
+gpuid_list="0"
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
-njob=4
+njob=1
 gpu_inference=true
 decode_cmd=utils/run.pl
 
@@ -20,7 +20,6 @@ if ${gpu_inference}; then
     inference_nj=$[${ngpu}*${njob}]
     _ngpu=1
 else
-    inference_nj=${njob}
     inference_nj=${njob}
     _ngpu=0
 fi
@@ -63,4 +62,4 @@ ${decode_cmd} --max-jobs-run "${inference_nj}" JOB=1:"${inference_nj}" "${_logdi
 
     for i in $(seq ${inference_nj}); do
         cat ${_logdir}/text.${i}
-    done | sort -k1 >${_dir}/text
+    done | sort -k1 >${_dir}/text

egs_modelscope/speechio/paraformer/paraformer_large_infer.sh

@@ -10,9 +10,9 @@ exp_dir=
 model_name=speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch
 model_revision="v1.0.4"     # please do not modify the model revision
 inference_nj=32
-gpuid_list="0,1" # set gpus, e.g., gpuid_list="0,1"
+gpuid_list="0" # set gpus, e.g., gpuid_list="0,1"
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
-njob=4  # the number of jobs for each gpu
+njob=1  # the number of jobs for each gpu
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 
 if ${gpu_inference}; then

egs_modelscope/wenetspeech/paraformer/paraformer_large_infer.sh

@@ -10,9 +10,9 @@ exp_dir=
 model_name=speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch
 model_revision="v1.0.4"     # please do not modify the model revision
 inference_nj=32
-gpuid_list="0,1" # set gpus, e.g., gpuid_list="0,1"
+gpuid_list="0" # set gpus, e.g., gpuid_list="0,1"
 ngpu=$(echo $gpuid_list | awk -F "," '{print NF}')
-njob=4  # the number of jobs for each gpu
+njob=1  # the number of jobs for each gpu
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 
 if ${gpu_inference}; then