Support bf16 type for transformer inference kernel to support Ds_Chat

intel / intel-extension-for-deepspeed

Intel® Extension for DeepSpeed* is an extension to DeepSpeed that brings feature support with SYCL kernels on Intel GPU(XPU) device. Note XPU is already supported in stock DeepSpeed (upstream).

MIT License

57 stars 19 forks source link

Support bf16 type for transformer inference kernel to support Ds_Chat #64

Closed ys950902 closed 10 months ago

ys950902 commented 10 months ago

Add the flag to support bf16 type for transformer inference kernel to enable Deepspeed-Chat

ys950902 commented 10 months ago

@baodii @rogerxfeng8 please review, to enable DeepSpeed-Chat.

baodii commented 10 months ago

LGTM