Minimal example for using ET and AOTI #763
Conversation
facebook-github-bot
added
the
CLA Signed
|
I would love to get your ideas, advice and perspectives on what to do with Andrej's original runner, enabled for DSO and PTE execution, in the long-term. I think this example is super instructive and useful because it answers the question "what is the minimum amount of work I have to do to support both DSO and PTE models from my own native application". In many ways, it's a lesser runner/run.cpp but instructive because of its bare-boneness, which is a mark of Andrej's genius to explaining this field to people. That being said, I'm struggling with defining the relationship of this example to the rest of torchchat and how to make this available to the public at large as an example, not a supported feature. (Nikita had suggested for another similar scenario to create an experimental directory to gather such examples.) I can see this being an example like we have in ExecuTorch but targeting server to mobile integration, or a PyTorch.org tutorial. Curious to get your thoughts and perspective. |
|
Thank you @iseeyuan ! |
* minimal example for enabling Andrej's runner, from commit 2d47702 * Minimal example
* minimal example for enabling Andrej's runner, from commit 2d47702 * Minimal example
* minimal example for enabling Andrej's runner, from commit 2d47702 * Minimal example
* minimal example for enabling Andrej's runner, from commit 2d47702 * Minimal example
* minimal example for enabling Andrej's runner, from commit 2d47702 * Minimal example
* minimal example for enabling Andrej's runner, from commit 2d47702 * Minimal example
Added to parking_lot from back-level unmodified pre-llama3 runner
Might move to torchchat/unsupported or to a separate PyTorch tutorial or repo
Advantageously, in torchchat we could run some CI, and have the infra to build models.
If we decide not to include it as unsupported example in torchchat, make it a separate tutorial?