---
title: zipformer-xlarge
canonical_url: "https://www.modelscope.cn/models/pkufool/zipformer-xlarge"
md_url: "https://www.modelscope.cn/models/pkufool/zipformer-xlarge.md"
repository: pkufool/zipformer-xlarge
last_updated: 2026-06-22
license: "Apache License 2.0"
pipeline_tag: auto-speech-recognition
tasks:
  - auto-speech-recognition
library_name:
  - onnx
  - pytorch
frameworks:
  - Pytorch
downloads: 10
stars: 1
---

# zipformer-xlarge

> zipformer-xlarge - pkufool 在 ModelScope 开源的模型。This is a xlarge zipformer model developed by Xiaomi AI Lab Next-gen-Kaldi team. The model was trained on around 20,0000 hours of open-sourced Chinese and English datasets. The number of parameters is around 300M.

pkufool/zipformer-xlarge 是 ModelScope 魔搭社区上的auto-speech-recognition模型，采用 Apache License 2.0 许可。

- **Repository**: pkufool/zipformer-xlarge
- **License**: Apache License 2.0
- **Tasks**: auto-speech-recognition
- **Downloads**: 10
- **Stars**: 1
- **Last updated**: 2026-06-22

Source: https://www.modelscope.cn/models/pkufool/zipformer-xlarge

---

## Introduction

This is a xlarge [zipformer](https://arxiv.org/pdf/2310.11230) model developed by Xiaomi AI Lab Next-gen-Kaldi team. The model was trained on around 20,0000 hours of open-sourced
Chinese and English datasets. The number of parameters is around 300M.

The performance on some popular test sets (CER for Chinese, WER for English).

> This model was trained with ctc head only.

| Head | aishell test 1 / 2 |  wenetspeech test-net/meetting | Common Voice zh | kespeech test | librispeech test-clean / other | gigaspeech test | Common voice en | tedium test |
| -- | -- | -- | -- | -- | -- | -- | -- | -- |
| CTC | 1.61 / 2.7  | 5.35 / 6.39 | 8.26 | 5.74 | 3.51 / 7.78 | 14.53 | 28.57 | 15.07 |

Please refer to [zipformer in github](https://github.com/pkufool/zipformer) for model details.

> Training set list: Librispeech, Gigaspeech, Commonvoice-2022(zh + en), Libriheavy, Emilia (zh+en), AIshell 2, Wenetspeech, Wenetspeech4tts, Kespeech, AIshell, aidatatang, aishell4, alimeeting, magicdata, primewords, stcmds, thchs30.


## Download model

### By SDK

```
# install modelscope
pip install modelscope

# download the model
from modelscope import snapshot_download
model_dir = snapshot_download('pkufool/zipformer-xlarge')
```

### By git
```
git clone https://www.modelscope.cn/pkufool/zipformer-xlarge.git
```


## Inference

### Install zipformer
```
pip install zipformer
```

### Examples (with downloaded model)

> You have to download the model by youself and provide the model path.

```
# jit script model
zipformer inference \
    --model zipformer-xlarge/jit_model.pt \
    --ctc 1 \
    --model-type jit \
    --tokens data/tokens.txt \
    data/en.wav data/zh.wav

# onnx model
zipformer inference \
    --model zipformer-xlarge/ctc.onnx \
    --ctc 1 \
    --model-type onnx \
    --tokens data/tokens.txt \
    data/en.wav data/zh.wav

# onnx fp16 model
zipformer inference \
    --model zipformer-xlarge/ctc.fp16.onnx \
    --ctc 1 \
    --model-type onnx \
    --tokens data/tokens.txt \
    data/en.wav data/zh.wav

# onnx int8 model
zipformer inference \
    --model zipformer-xlarge/ctc.int8.onnx \
    --ctc 1 \
    --model-type onnx \
    --tokens data/tokens.txt \
    data/en.wav data/zh.wav
```

### Example (with model name in modelscope)

> You don't have to download the model, the model will be downloaded automatically from modelscope.

```
# jit script model
zipformer inference \
    --ms-model pkufool/zipformer-xlarge \
    --ctc 1 \
    --model-type jit \
    data/en.wav data/zh.wav

# onnx model
zipformer inference \
    --ms-model pkufool/zipformer-xlarge \
    --ctc 1 \
    --model-type onnx \
    data/en.wav data/zh.wav

# onnx fp16 model
zipformer inference \
    --ms-model pkufool/zipformer-xlarge \
    --ctc 1 \
    --dtype fp16 \
    --model-type onnx \
    data/en.wav data/zh.wav

# onnx int8 model
zipformer inference \
    --ms-model pkufool/zipformer-xlarge \
    --ctc 1 \
    --dtype int8 \
    --model-type onnx \
    data/en.wav data/zh.wav
```

## Training

### Install zipformer
```
pip install zipformer
```

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

zipformer train \
    --world-size 8 \
    --exp-dir zipformer/exp_xlarge \
    --num-encoder-layers 2,2,4,5,4,2 \
    --feedforward-dim 512,1024,2048,3072,2048,1024 \
    --encoder-dim 192,384,768,1024,768,384 \
    --encoder-unmasked-dim 192,256,320,512,320,256 \
    --bpe-model zh-en-yue-11661 \
    --training-sets data/training_set.lst
    --num-epochs 20 \
    --use-fp16 1 \
    --start-epoch 1 \
    --use-cr-ctc 1 \
    --use-ctc 1 \
    --base-lr 0.035 \
    --use-transducer 0 \
    --use-attention-decoder 0 \
    --enable-spec-aug 0 \
    --cr-loss-scale 0.2 \
    --time-mask-ratio 2.5 \
    --lr-hours 100000 \
    --num-workers 2 \
    --max-duration 600
```

## Export model

### Install zipformer
```
pip install zipformer
```

```
zipformer export \
    --use-ctc 1 \
    --use-transducer 0 \
    --num-encoder-layers 2,2,4,5,4,2 \
    --feedforward-dim 512,1024,2048,3072,2048,1024 \
    --encoder-dim 192,384,768,1024,768,384 \
    --encoder-unmasked-dim 192,256,320,512,320,256 \
    --exp-dir zipformer/exp_xlarge \
    --bpe-model zh-en-yue-11661 \
    --iter ITER \
    --avg AVG
```

## Citation

```
@inproceedings{yao2024zipformer,
  title={Zipformer: A faster and better encoder for automatic speech recognition},
  author={Yao, Zengwei and Guo, Liyong and Yang, Xiaoyu and Kang, Wei and Kuang, Fangjun and Yang, Yifan and Jin, Zengrui and Lin, Long and Povey, Daniel},
  booktitle={International Conference on Learning Representations},
  volume={2024},
  pages={44440--44455},
  year={2024}
}

@inproceedings{yao2025cr,
  title={Cr-ctc: Consistency regularization on ctc for improved speech recognition},
  author={Yao, Zengwei and Kang, Wei and Yang, Xiaoyu and Kuang, Fangjun and Guo, Liyong and Zhu, Han and Jin, Zengrui and Li, Zhaoqing and Lin, Long and Povey, Daniel},
  booktitle={International Conference on Learning Representations},
  volume={2025},
  pages={26850--26868},
  year={2025}
}
```