diff --git a/egs/aishell/paraformer/run.sh b/egs/aishell/paraformer/run.sh
index 06322ce77..c5ee1f289 100755
--- a/egs/aishell/paraformer/run.sh
+++ b/egs/aishell/paraformer/run.sh
@@ -8,7 +8,7 @@ gpu_num=2
 count=1
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 # for gpu decoding, inference_nj=ngpu*njob; for cpu decoding, inference_nj=njob
-njob=8
+njob=1
 train_cmd=utils/run.pl
 infer_cmd=utils/run.pl
 
@@ -219,7 +219,7 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
         fi
         ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \
             python -m funasr.bin.asr_inference_launch \
-                --batch_size 1 \
+                --batch_size 100 \
                 --ngpu "${_ngpu}" \
                 --njob ${njob} \
                 --gpuid_list ${gpuid_list} \
diff --git a/egs/aishell/paraformerbert/run.sh b/egs/aishell/paraformerbert/run.sh
index 6f331ec43..a39ac7758 100755
--- a/egs/aishell/paraformerbert/run.sh
+++ b/egs/aishell/paraformerbert/run.sh
@@ -8,7 +8,7 @@ gpu_num=2
 count=1
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 # for gpu decoding, inference_nj=ngpu*njob; for cpu decoding, inference_nj=njob
-njob=8
+njob=1
 train_cmd=utils/run.pl
 infer_cmd=utils/run.pl
 
@@ -235,7 +235,7 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
         fi
         ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \
             python -m funasr.bin.asr_inference_launch \
-                --batch_size 1 \
+                --batch_size 100 \
                 --ngpu "${_ngpu}" \
                 --njob ${njob} \
                 --gpuid_list ${gpuid_list} \
diff --git a/funasr/bin/asr_inference_paraformer.py b/funasr/bin/asr_inference_paraformer.py
index 15a37f7da..9dcd0b8b9 100755
--- a/funasr/bin/asr_inference_paraformer.py
+++ b/funasr/bin/asr_inference_paraformer.py
@@ -441,7 +441,7 @@ def inference(
             "decoding, feature length: {}, forward_time: {:.4f}, rtf: {:.4f}".
                 format(length, forward_time, 100 * forward_time / (length*lfr_factor)))
         
-        for batch_id in range(len(results)):
+        for batch_id in range(_bs):
             result = [results[batch_id][:-2]]
     
             key = keys[batch_id]
diff --git a/funasr/models/predictor/cif.py b/funasr/models/predictor/cif.py
index 2eba4e251..1d196a7ce 100644
--- a/funasr/models/predictor/cif.py
+++ b/funasr/models/predictor/cif.py
@@ -31,10 +31,12 @@ class CifPredictor(nn.Module):
         alphas = torch.sigmoid(output)
         alphas = torch.nn.functional.relu(alphas * self.smooth_factor - self.noise_threshold)
         if mask is not None:
-            alphas = alphas * mask.transpose(-1, -2).float()
+            mask = mask.transpose(-1, -2).float()
+            alphas = alphas * mask
         if mask_chunk_predictor is not None:
             alphas = alphas * mask_chunk_predictor
         alphas = alphas.squeeze(-1)
+        mask = mask.squeeze(-1)
         if target_label_length is not None:
             target_length = target_label_length
         elif target_label is not None: