mirror of https://github.com/kubeflow/trainer.git
* feat(runtimes): Support Distributed MLX on CUDA Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Remove arm build from MLX runtime Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Update get_runtime_packages API Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Force to change vars in examples Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Remove LD_LIBRARY_PATH updates Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Add patch command to DeepSpeed example Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Cleanup apt packages Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> * Reduce MLX and DeepSpeed image size Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> --------- Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com> |
||
|---|---|---|
| .. | ||
| initializers | ||
| runtimes | ||
| trainer-controller-manager | ||
| trainers/torchtune | ||