pkufool's picture
Update README.md
7818cee
---
license: apache-2.0
---
See https://github.com/k2-fsa/icefall/pull/454
### training command:
```bash
./pruned_transducer_stateless5/train.py \
--exp-dir pruned_transducer_stateless5/exp \
--num-encoder-layers 18 \
--dim-feedforward 2048 \
--nhead 8 \
--encoder-dim 512 \
--decoder-dim 512 \
--joiner-dim 512 \
--full-libri 1 \
--dynamic-chunk-training 1 \
--causal-convolution 1 \
--short-chunk-size 20 \
--num-left-chunks 4 \
--max-duration 300 \
--world-size 4 \
--start-epoch 1 \
--num-epochs 25
```
You can find the tensorboard log here <https://tensorboard.dev/experiment/rO04h6vjTLyw0qSxjp4m4Q>
### The decoding command is:
```bash
decoding_method="greedy_search" # "fast_beam_search", "modified_beam_search"
for chunk in 2 4 8 16; do
for left in 32 64; do
./pruned_transducer_stateless5/decode.py \
--num-encoder-layers 18 \
--dim-feedforward 2048 \
--nhead 8 \
--encoder-dim 512 \
--decoder-dim 512 \
--joiner-dim 512 \
--simulate-streaming 1 \
--decode-chunk-size ${chunk} \
--left-context ${left} \
--causal-convolution 1 \
--epoch 25 \
--avg 5 \
--exp-dir ./pruned_transducer_stateless5/exp \
--max-sym-per-frame 1 \
--max-duration 1000 \
--decoding-method ${decoding_method}
done
done
```
### export command is:
```bash
./pruned_transducer_stateless5/export.py \
--streaming-model 1 \
--causal-convolution 1 \
--num-encoder-layers 18 \
--dim-feedforward 2048 \
--nhead 8 \
--encoder-dim 512 \
--decoder-dim 512 \
--joiner-dim 512 \
--epoch 25 \
--avg 5 \
--exp-dir ./pruned_transducer_stateless5/exp
./pruned_transducer_stateless5/export.py \
--streaming-model 1 \
--causal-convolution 1 \
--num-encoder-layers 18 \
--dim-feedforward 2048 \
--nhead 8 \
--encoder-dim 512 \
--decoder-dim 512 \
--joiner-dim 512 \
--epoch 25 \
--avg 5 \
--exp-dir ./pruned_transducer_stateless5/exp \
--jit 1
```