pip install executorch doesn't install llama custom ops

[mergennachin]

🚀 The feature, motivation and pitch

pip install executorch doesn't install sdpa_with_kv_cache op

Currently, llama is such a first-class citizen and so it makes sense to install these important custom ops by default as part of the pip installation.

Here's a repro step:

https://colab.research.google.com/drive/1xofI6KV1a5DX1Zlu2YONu92a9FgBPHSB#scrollTo=f9r__rm15mbp

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

[larryliu0820]

This only repros in colab. I tried 0.5.0.dev20241126 nightly the following works:

from executorch.kernels import quantized

But importing llm custom op failed

from executorch.extension.llm.custom_ops import sdpa_with_kv_cache

[larryliu0820]

#7096 should fix

[mergennachin]

Hi @larryliu0820

It's still having same issue. See repro using executorch==0.5.0.dev20241201

https://colab.research.google.com/drive/1xofI6KV1a5DX1Zlu2YONu92a9FgBPHSB#scrollTo=f9r__rm15mbp