Wei Kang
Wei Kang
Fix https://github.com/danpovey/fast_rnnt/issues/10 and update the method to generate pruning bounds. See also in k2 repository https://github.com/k2-fsa/k2/pull/1009
The demo apk is uploaded to [huggingface](https://huggingface.co/pkufool/next-gen-kaldi-android), the decoding method is "greedy_search", will add graph based decoding method in next PR.
Fix #14
子目录部署问题
非常精致小巧的评论系统!感谢作者。 我有一个问题,我用nginx将twikoo 部署到 `https://www.xxx.com/twikoo` 上面,已经显示云端部署成功: ``` { "code": 100, "message": "Twikoo 云函数运行正常,请参考 https://twikoo.js.org/quick-start.html#%E5%89%8D%E7%AB%AF%E9%83%A8%E7%BD%B2 完成前端的配置", "version": "1.6.7" } ``` 可是在页面使用时,总是报评论失败的错误: ``` Twikoo: 评论失败 Error: Twikoo 云函数运行正常,请参考 https://twikoo.js.org/quick-start.html#%E5%89%8D%E7%AB%AF%E9%83%A8%E7%BD%B2 完成前端的配置 ``` 之前我部署在 `http://twikoo.xxx.com`...
The purpose of this PR is to fix the `default_context.cu` so that we can use k2 without Pytorch. There are still some issues to be fixed. - [ ] `NvToolExt.h`...
Implement fast_emit proposed in https://arxiv.org/pdf/2010.11148.pdf
This PR implements the levenshtein distance on GPU, it can run in batches and has boundary support. From a simple benchmark as follows, you can get quite a lot speedup...
Some fixes to support fast_beam_search in sherpa.