Downloads · 30 days
0
mmm678/icefall_asr_aishell_pruned_transducer_stateless7_bbpe
icefall_asr_aishell_pruned_transducer_stateless7_bbpe is a machine learning model from mmm678. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Downloads · 30 days
0
Access
Public
Updated Apr 3, 2026
Repo size
733 MB
Likes
0
Public
Click a slice to open those files.
.pt732 MB · 99%
From the Hugging Face model README
The results:
| Vocab size | Greedy search(dev & test) | Modified beam search(dev & test) | Fast beam search (dev & test) | Fast beam search LG (dev & test) | comments |
|---|---|---|---|---|---|
| 500 | 4.31 & 4.59 | 4.25 & 4.54 | 4.27 & 4.55 | 4.07 & 4.38 | --epoch 48 --avg 29 |
The training command:
export CUDA_VISIBLE_DEVICES="4,5,6,7"
./pruned_transducer_stateless7_bbpe/train.py \
--world-size 4 \
--num-epochs 50 \
--start-epoch 1 \
--use-fp16 1 \
--max-duration 800 \
--bpe-model data/lang_bbpe_500/bbpe.model \
--exp-dir pruned_transducer_stateless7_bbpe/exp \
--lr-epochs 6 \
--master-port 12535
The decoding command:
for m in greedy_search modified_beam_search fast_beam_search fast_beam_search_LG; do
./pruned_transducer_stateless7_bbpe/decode.py \
--epoch 48 \
--avg 29 \
--exp-dir ./pruned_transducer_stateless7_bbpe/exp \
--max-sym-per-frame 1 \
--ngram-lm-scale 0.25 \
--ilme-scale 0.2 \
--bpe-model data/lang_bbpe_500/bbpe.model \
--max-duration 2000 \
--decoding-method $m
done