mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-08 09:32:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update Zipformer-large result on LibriSpeech (#1343)

Browse Source 
* update zipformer-large result on librispeech
This commit is contained in:
Zengwei Yao 2023-10-26 17:35:12 +08:00 committed by GitHub
parent 770c495484
commit c0a53271e2
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 58 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
README.md
View File

@ -118,11 +118,12 @@ We provide a Colab notebook to run a pre-trained transducer conformer + stateles
#### k2 pruned RNN-T
| Encoder | Params | test-clean | test-other |
|-----------------|--------|------------|------------|
| zipformer | 65.5M | 2.21 | 4.79 |
| zipformer-small | 23.2M | 2.42 | 5.73 |
| zipformer-large | 148.4M | 2.06 | 4.63 |
| Encoder | Params | test-clean | test-other | epochs | devices |
|-----------------|--------|------------|------------|---------|------------|
| zipformer | 65.5M | 2.21 | 4.79 | 50 | 4 32G-V100 |
| zipformer-small | 23.2M | 2.42 | 5.73 | 50 | 2 32G-V100 |
| zipformer-large | 148.4M | 2.06 | 4.63 | 50 | 4 32G-V100 |
| zipformer-large | 148.4M | 2.00 | 4.38 | 174 | 8 80G-A100 |
Note: No auxiliary losses are used in the training and no LMs are used
in the decoding.

52
egs/librispeech/ASR/RESULTS.md
View File

@ -245,6 +245,58 @@ for m in greedy_search modified_beam_search fast_beam_search; do
done
```
##### large-scaled model, number of model parameters: 148439574, i.e., 148.4 M, trained on 8 80G-A100 GPUs
The tensorboard log can be found at
<https://tensorboard.dev/experiment/95TdNyEuQXaWK2PzFpD9yg/>
You can find a pretrained model, training logs, decoding logs, and decoding results at:
<https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-large-2023-10-26-8-a100>
You can use <https://github.com/k2-fsa/sherpa> to deploy it.
| decoding method | test-clean | test-other | comment |
|----------------------|------------|------------|-----------------------|
| greedy_search | 2.00 | 4.47 | --epoch 174 --avg 172 |
| modified_beam_search | 2.00 | 4.38 | --epoch 174 --avg 172 |
| fast_beam_search | 2.00 | 4.42 | --epoch 174 --avg 172 |
The training command is:
```bash
export CUDA_VISIBLE_DEVICES="0,1,2,3"
./zipformer/train.py \
--world-size 8 \
--num-epochs 174 \
--start-epoch 1 \
--use-fp16 1 \
--exp-dir zipformer/exp-large \
--causal 0 \
--num-encoder-layers 2,2,4,5,4,2 \
--feedforward-dim 512,768,1536,2048,1536,768 \
--encoder-dim 192,256,512,768,512,256 \
--encoder-unmasked-dim 192,192,256,320,256,192 \
--full-libri 1 \
--max-duration 2200
```
The decoding command is:
```bash
export CUDA_VISIBLE_DEVICES="0"
for m in greedy_search modified_beam_search fast_beam_search; do
./zipformer/decode.py \
--epoch 174 \
--avg 172 \
--exp-dir zipformer/exp-large \
--max-duration 600 \
--causal 0 \
--decoding-method $m \
--num-encoder-layers 2,2,4,5,4,2 \
--feedforward-dim 512,768,1536,2048,1536,768 \
--encoder-dim 192,256,512,768,512,256 \
--encoder-unmasked-dim 192,192,256,320,256,192
done
```
#### streaming
##### normal-scaled model, number of model parameters: 66110931, i.e., 66.11 M