ivila commented on PR #173: URL: https://github.com/apache/incubator-teaclave-trustzone-sdk/pull/173#issuecomment-2720254757
> Other recommendations: > > * Seems in `ta/` only have `crates/` and a `Cargo.toml`, how about removing the `crates/` directory layer and simplifying the structure to `mnist-rs/ta/inference`, `mnist-rs/ta/training`, etc.? > * How about moving this example to `projects/`, which contains real-world and more specific scenario-based examples? Prefer `projects/[scenario/field]/mnist`, similar to `projects/web3/eth-wallet`. For the two recommendations. > how about removing the `crates/` directory layer and simplifying the structure? done > How about moving this example to `projects/`? I don’t think we should😂. This example only includes a simple toy model with just a Linear layer, which isn’t a proper model. For comparison, a typical CNN model with 768 inputs and 10 outputs would require at least 300MB of memory for training, while our QEMU8 image only has 20MB of memory allocated for the TA. Additionally, a real-world project would be much more complex. For example, the model crate would be automatically generated to meet different requirements, such as varying shapes or custom image-to-tensor methods. This example is meant to demonstrate that we can perform deep learning inside TEE, and I think that’s sufficient for now. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: dev-unsubscr...@teaclave.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@teaclave.apache.org For additional commands, e-mail: dev-h...@teaclave.apache.org
