Yao Matrix comments

Results 18 comments of


                                            Yao Matrix

apply amp bf16 on textual inversion

@kding1 this is the PR we mentioned, to upstream BF16 common practices to textual inversion, which will benefit Intel 4th generation Xeon platform also.

add intel xpu support for TGI

@Narsil @mfuntowicz , pls feel free to let us know any issues, since CI is ready, I think we can launch CI and merge this PR, thx. cc @kding1

Enable traced model for text-generation task

Yes, echo. W/ PyTorch 2.0 introduced, suppose we will see more and more performance benefit out of jit for deployment.

fix prompt tunning + deepspeed zero3 + checkpoint_saving hang issue

@pacman100 , @ArthurZucker what's your suggestion to move on for this issue? any comments on Yi's proposal?

Enable traced model for text-generation task

@sgugger @gante

[RFC]Integrate Intel Extension for Pytorch into accelerate to get out-of-box optimizations on intel platform.

@pacman100 , @jianan-gu, @sywangyi ,w/ IPEX load_state_dict support, do we see it's OK to merge this PR?

[RFC]Integrate Intel Extension for Pytorch into accelerate to get out-of-box optimizations on intel platform.

@kding1

feat: enable loading tokenizer from YAML file for models not supported in transformers

but there is always models which not supported by local HF APIs and need remote execution, I think we still need these knobs here so Haystack can always support the...

Allreduce cpu example fails with CCL_WORKER_COUNT > 1

@piotrchmiel , you can try this. `echo 0 > /proc/sys/kernel/yama/ptrace_scope`.

Enable intel devices CPU/XPU/HPU for python backend

@OlivierDehaene , could you help review? Thx