-
Hi, all!
Trying to learn a Squeezeformer-S using bf16 precision, but I got **~15 s/it**. But everything is fine learning Conformer with the same number of params with bf16 (got **~3 s/it**) . Also ev…
-
# 一、问题描述
- 基于SqueezeFormer的训练模型ONNX转换时,遭遇MatMul_2995问题:“2022-11-21 14:51:01.052900711 [E:onnxruntime:, sequential_executor.cc:369 Execute] Non-zero status code returned while running MatMul node. Nam…
-
# 一、问题描述
- 感谢修复https://github.com/wenet-e2e/wenet/issues/1569#issue-1457497469;
- 但、基于SqueezeFormer的ONNX模型转换后,结果存在不一致问题,具体如下:
- 模型转换前、wer评估的抽样结果:
![image](https://user-images.githubusercontent.c…
-
Just pulled the latest wenet code and tried out Squeezformer. The training is failed with this log attached below. Any suggestion would be helpful. Thanks.
`the number of model params: 135,220,418…
-
**Is your feature request related to a problem? Please describe.**
Squeezeformer arcitecture support was added in `nemo_toolkit==1.12.0`
But looks like huggingface docker has `nemo_toolkit==1.10.0`
…
-
# ❓ Questions & Help
## Details
안녕하세요 Openspeech를 처음 사용해보는 학생입니다.
Custom Dataset으로 학습을 해보고자 Issue와 Langcon 영상을 참고해서, preprocess와 tokenizer 부분을 따로 구현한 뒤, 가이드를 참고해 hydra_train.py를 실행했더니, 제가 추가한…
-
Squeezeformer looks promising and uses some of the reusable components from conformer backbone. Creating a placeholder for discussion and implementation.
[Official Implementation](https://github.co…
-
微博内容精选