sanchitintel

> In this case, the number of outputs from `split` is dynamic, how's this represented in the framework graph? @ZhennanQin, currently, the IPEX/PyTorch JIT framework graph is created by tracing...

> * If it is not a Constant according to the TorchScript IR, we're not able to map it to the static attr. Another runtime representation of `split_size` will be...

Hi @Xia-Weiwen, can we wait for a couple of days, and choose another commit instead? Thanks!

> So if I have to provide alternate implementations of the unsupported partitions, is there a way to integrate those alternate implementations with those that are supported Let's say you...

I noticed that even if Attention mask is absent ([it may or may not be present in GPT2](https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt2/modeling_gpt2.py#L204-L206)), pattern-matching & replacing GPT2 SDPA computation (I can see with a breakpoint...

Hi! I read your blog-post on GPT2 performance with the SDPA op, using the HF Optimum library. I was mistaken about GPT2 not being able to use `is_causal` as `True`...

> tested both a hyperthreaded multicore and non hyperthreaded multicore. It gives the correct thread count now. Hi @gajjanag, can you please clarify if you meant you were getting incorrect...

Thanks for confirming, @gajjanag! That hasn't been my experience, with HyperThreading enabled (without this patch) -

Hi @jansel @eellison, in this PR, I'm trying to add 1. Integration code for oneDNN Graph fusions to work with Inductor (disabled by default) 2. A fusion pattern to verify...

Hi @jansel, thanks for your feedback! In the future, oneDNN Graph would also support XPU (Intel GPU). > It creates bifurcation between CPU/GPU backends which will be hard to maintain....