Implemented SGD optimizer with momentum and added tests. #217
Conversation
This file contains hidden or 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
Muxas
requested changes
Mar 31, 2025
| @@ -34,6 +34,7 @@ | |||
| #include <nntile/kernel/prod_inplace.hh> | |||
| #include <nntile/kernel/randn.hh> | |||
| #include <nntile/kernel/relu.hh> | |||
| #include <nntile/kernel/sgd_momentum.hh> | |||
| * distributed-memory heterogeneous systems based on StarPU runtime system. | ||
| * | ||
| * @file include/nntile/kernel/sgd_momentum.hh | ||
| * SGD_MOMENTUM low-level kernels |
|
|
||
| // SGD_MOMENTUM operation on a buffer | ||
| template<typename T> | ||
| void cpu(Index nelems, T *data) |
There was a problem hiding this comment.
Totally incorrect API, where are the gradients, parameters like learning rate?
| { | ||
|
|
||
| template<typename T> | ||
| void cuda(cudaStream_t stream, Index nelems, T *data) |
There was a problem hiding this comment.
Parameters of the function are incorrect
| @@ -127,10 +128,11 @@ void init() | |||
| gemm::init(); | |||
| hypot::init(); | |||
| hypot_scalar_inverse::init(); | |||
| nrm2::init(); | |||
| nrm2::init(); | |||
There was a problem hiding this comment.
do not put binary files into git repo
There was a problem hiding this comment.
do not put binary files into git repo
There was a problem hiding this comment.
tests shall appear in test directory
| # NNTile is software framework for fast training of big neural networks on | ||
| # distributed-memory heterogeneous systems based on StarPU runtime system. | ||
| # | ||
| # @file wrappers/python/nntile/optimizer/sgd.py |
There was a problem hiding this comment.
take a look at wrappers/python/optimizer/adam.py
| @@ -516,6 +516,15 @@ void def_mod_tensor(py::module_ &m) | |||
| m.def("relu_fp32", &relu<fp32_t>); | |||
| m.def("relu_fp32_fast_tf32", &relu<fp32_fast_tf32_t>); | |||
|
|
|||
| // Add activation functions for Tensor<T> | |||
| m.def("sgd_momentum_async_fp64", &sgd_momentum_async<fp64_t>); | |||
There was a problem hiding this comment.
This function call shall be propagated also into wrappers/python/nntile/functions.py (find adam_step)
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.
I implemented the SGD optimizer with momentum and updated local repository at the kernel, starpu and tensor levels. I also provided tests for the sgd with momentum on the home directory to verify and check that the optimizer works correctly