[docs] kernels

[stevhliu]

adds a kernels section in the Accelerate inference docs with the results:

• cross-linked to Attention backends docs which demonstrates support for loading attention kernels with set_attention_backend
• defer to the blog post and pipeline integration guide for more details about implementing non-attention kernels since this is more involved and already well-documented there

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[sayakpaul]

Thanks a lot for prioritizing it.

[sayakpaul]

The kernel skill basically lets users get an agent to write custom kernels for a model and hardware. It's not specific to the attention processor but also other modules as well such RMSNorm. Should we make it clearer?

[stevhliu]

lmk if this is clearer!

[sayakpaul]

It wasn't just RMSNorm but also other modules implemented with custom kernels.

[stevhliu]

i mention RMSNorm as an example only for the benchmark results below

[sayakpaul]

Thanks! I would like to also see what @burtenshaw thinks about this.