-
|
Am wondering if you guys would be interested in contributions of implementations of layers and optimizers in C++? Reason is otherwise people who want to use layers or optimizers will either need to stick with Python (as you guys have implemented it in Python) or they have to implement it themselves which might hurt the barrier to entry. |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 1 reply
-
|
May fix the interference
…On Sat, Mar 2, 2024, 9:11 PM Cemlyn ***@***.***> wrote:
Am wondering if you guys would be interested in contributions of
implementations of layers and optimizers in C++?
Reason is otherwise people who want to use layers or optimizers will
either need to stick with Python (as you guys have implemented it in
Python) or they have to implement it themselves which might hurt the
barrier to entry.
—
Reply to this email directly, view it on GitHub
<#774>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/BGNW7PRQQ3FBCDGVD66M6TLYWKIF7AVCNFSM6AAAAABEDTDBT2VHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZWGMYDQNBZHE>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
It's something we've discussed adding in the past, but haven't quite found enough of a reason to do it. At the moment we're not planning to add that level in C++ until there is enough demand for it (as it adds another surface for us to maintain). FWIW we also have MLX Swift which has all the high-level layers + otpimizers: https://github.com/ml-explore/mlx-swift/. I'm not sure what your goals are but maybe that would be a good alternative. However, if this is something super useful for you, I would encourage you to start a separate project which builds some of those layers out on top of MLX in C++. That would make it 1. more concrete so we can see how it looks and if it makes sense to include and 2. see if it is used enough to merit pulling into core. |
Beta Was this translation helpful? Give feedback.
-
|
100% makes sense, for now I'll try porting some of the layers and optimizers that I need for my project. I'm not a C++ expert but I'll be giving it a crack over the course of this week. Depending on how it goes I might consider creating a separate project for hosting some implemented layers and optimizers! Worse case my project can be a little example for others. Thank you and hope you have a good weekend! |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the info
…On Sun, Mar 3, 2024, 7:38 AM Cemlyn ***@***.***> wrote:
100% makes sense, for now I'll try porting some of the layers and
optimizers that I need for my project <https://github.com/cemlyn007/a3c>.
I'm not a C++ expert but I'll be giving it a crack over the course of this
week. Depending on how it goes I might consider creating a separate project
for hosting some implemented layers and optimizers! Worse case my project
can be a little example for others.
Thank you and hope you have a good weekend!
—
Reply to this email directly, view it on GitHub
<#774 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/BGNW7PSKDBWV4O3NDYWJDMTYWMRWHAVCNFSM6AAAAABEDTDBT2VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4DMNJXGU4DA>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
It's something we've discussed adding in the past, but haven't quite found enough of a reason to do it. At the moment we're not planning to add that level in C++ until there is enough demand for it (as it adds another surface for us to maintain).
FWIW we also have MLX Swift which has all the high-level layers + otpimizers: https://github.com/ml-explore/mlx-swift/. I'm not sure what your goals are but maybe that would be a good alternative.
However, if this is something super useful for you, I would encourage you to start a separate project which builds some of those layers out on top of MLX in C++. That would make it 1. more concrete so we can see how it looks and if it makes sense to inc…