chore(mlx-c): update to 0.0.9 #105
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
minghuaw
pushed a commit
that referenced
this pull request
Sep 9, 2024
minghuaw
added a commit
that referenced
this pull request
Oct 20, 2024
* add Module and Sequential * added some activation fn * added compiled activation functions * removed swift references * added more activation functions * added docs * added remaining activation functions * renamed clone to deep_clone to avoid confusion and changed to public * changed Module trait to take ref * cargo fmt * make argument_numbers optional * improve doc * use ValueAndGrad trait instead of just value_and_grad fn * remove ValueAndGrad, use value_and_grad and value_and_grad_with_payload instead * add docs * Module::update is complicated by ownership * Trust synchronization on the cpp side (#102) * remove mut ref and smart ptr on rust side * use clone instead of mlx_retain * removed OwnedOrRef * changed Module trait to take ref * cargo fmt * use ValueAndGrad trait instead of just value_and_grad fn * remove ValueAndGrad, use value_and_grad and value_and_grad_with_payload instead * change OwnedOrRef to Cow * Change to ValueAndGrad trait to allow owned closure * added nested * change generic T to V * moved nested from mlx-rs to mlx-nn * initial impl of value_and_grad with HashMap * use generic instead of hard coded HashMap * use Rc<str> to avoid expensive clone * impl value_and_grad_with_hashmap * initial impl of model_value_and_grad * added mlx-nn-module crate and ModuleParameters derive macro * cargo clippy, fix, & fmt * move closure inner into the returned closure * added as_trainable_nested_value * initial attempt at impl Linear and Bilinear * tested value_and_grad * change setter fn for optional fields to have "with_" prefix * derive ModuleParameters for Sequential * cargo clippy, fix, fmt * Trust synchronization on the cpp side (#102) * remove mut ref and smart ptr on rust side * use clone instead of mlx_retain * removed OwnedOrRef * chore(example): add automatic_differentiation example (#103) * chore(mlx-c): update to 0.0.9 (#105) * cargo clippy, fix, fmt * add example in ModuleParameters doc * fixed error in doc example * added Conv2d and Conv3d * cargo fmt * added train() to Module trait & added docs to mlx-nn-module * wrap number literals in `array!()` to avoid SIGSEGV * added remaining docs * cargo fmt * impl SGD * added unit test for Sgd * change setters for optional configs * rename Sgd to SGD * Revert "rename Sgd to SGD" This reverts commit dbf6052. * added RmsProp optimizer * rename Module::train to Module::training_mode to avoid confusion * rename mod sequential to mod container * added panics section in doc * add mlx_nn::error::Error * add dropout mod * cargo clippy, fix, fmt * add unit tests for dropout * ported convolution unit tests from the swift binding * ported unit tests for linear * fixed bug and ported unit tests for activation modules * fix sigsegv caused by mixing borrow and owned value * cargo fmt * fix problem that was caused by clippy --fix * remove lint attr because the cause is already resolved * Use custom c shim for fallible closure (#116) * add support for closure with error in grads * compiled fn that should fail somehow doesn't fail after first failure * Revert "compiled fn that should fail somehow doesn't fail after first failure" This reverts commit 623ced1. * cargo clippy, fix, fmt * returns an empty mlx_vector_array to avoid invalid memory ref * propagate err with c binding fallible closure * compile fallible fn * rename *_with_error to *_fallible * cargo fmt * Restrict use of ScalarOrArray trait to operator impl and ClipBound (#108) * restrict use of ScalarOrArray trait to operator impl and ClipBound * fix doctest * Allow using named `shape` parameter in `array!()` macro (#109) * allow using array!() with shape * cargo fmt * fix error in unit test * simple c shim fails * value_and_grad is getting memory error even for success case * use custom shim c code for fallible closure * change fallible suffix to prefix and added fallible_value_and_grad_with_hashmap * revert back to use Exception for mlx-nn errors * use fallible version of compile * add cross_entropy * added binary_cross_entropy * added remaining losses * added losses unit tests * generate builder impl with option_builder proc macro * unified regular and fallible value_and_grad * moved mlx-nn-module into mlx-rs * initialize mnist example crate * change Exception::from to Excepion::custom * renamed mod optimizer to mod optimizers and do not re-export losses * change eval to take ref instead of mut ref * fix lifetime issue * initial nn example * suppress clippy::module_inception * cargo fmt * added eval_params and async_eval_params * allow taking ref of options * added smoke tests * remove commented code * added missing `#[option_builder]` * new attempt at builder pattern on Module * attempt new optional args handling on cross entropy * attemp builder pattern on RmsProp * impl basic generate_builder * moved to builder pattern for activation, losses and optimizer * removed WithBias and added builder pattern for conv and linear * cargo clippy and fmt * added builder for dropout modules * cargo clippy and fmt * removed unused generic builder * fixed wrong default due to derive * fix error caused by mixing operator and arith functions * generate Default impl if no mandatory field & hide internal macro from doc * added doc for GenerateBuilder macro * moved internal macros into separate crate --------- Co-authored-by: David Chavez <david@dcvz.io>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Updated version of #88