Try zipformer for the wenetspeech dataset

[csukuangfj]

Training command:

export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"

./pruned_transducer_stateless7/train.py \
  --world-size 8 \
  --num-epochs 20 \
  --use-fp16 1 \
  --max-duration 250 \
  --exp-dir pruned_transducer_stateless7/exp \
  --feedforward-dims  "512,512,1024,1024,512" \
  --master-port 12535 \
  --training-subset L