Dinghao Zhou

Results 41 issues of Dinghao Zhou

Open this to track different decoding strategies of the transducer model. I will update some materials and code snippets here Suggestions welcome and hope to collaborate

enhancement
help wanted

libtorch model cause memory growth for courrent call.

Greet job for gopher! Any plan to support distributed training?

will rnnt loss support fastemit?

triaged

latex转word 支持怎么样? 学院要求评阅版论文上传word , 图书馆要求pdf。有两种方案: 1 直接复制粘贴到word上, 公式较多,比较费时 2 pabdoc latex 转word 不知道支持效果怎么样

Is it possible to output token level timestep? eg: hello 100-600 world 712-900 .......

enhancement

这个pr 会将以下llm的组件引入wenet (llama gemma等都用了) - [x] gated mlp https://github.com/wenet-e2e/wenet/pull/2395 - [x] rope - [x] linear bias args https://github.com/wenet-e2e/wenet/pull/2394 - [x] multi query - [x] rms norm https://github.com/wenet-e2e/wenet/pull/2396 https://github.com/wenet-e2e/wenet/pull/2397 TODO -...

如果encoder没开ckpt,decoder开了ckpt, 会报错 但是这样很丑,如果有其他模块用了ckpt 也得加一行 感觉这里需要重构下

https://github.com/wenet-e2e/wenet/pull/2351#discussion_r1500361138 这里直接给decoder一个参数,