mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
74280d0913
A single-device version of Muon. Algorithm refers Keller Jordan's [Muon blogpost](https://kellerjordan.github.io/posts/muon/), and optionally incorporates [Moonshot's](https://github.com/MoonshotAI/Moonlight/blob/master/Moonlight.pdf) learning rate adjustment strategy. This implementation maintains a minimalist API and is consistent with other optimizer conventions. PyTorch team prefers to handle parameter filtering at a higher level, with the Muon optimizer performing only the msign computation for orthogonalization on all parameters it receives. Users are responsible for grouping parameters for different optimizers as needed. An example usage is shown below, and a more detailed example will be added to the [PyTorch examples](https://github.com/pytorch/examples) directory. **Usage** ```python model = MyModelForCausalLM # filter out your params manually muon_params = [...] adamw_params = [...] muon = Muon( params = muon_params lr=lr, wd=wd, ) adamw = AdamW( params = adamw_params lr=lr, wd=wd, ) # in training loop loss = model(input) loss.backward() muon.step() adamw.step() muon.zero_grad() adamw.zero_grad() ``` ~~**Additional usage**~~ ~~Users are also able to pass in self-defined `msign` function for orthogonalization, and learning rate adjustment function. Interface defined below:~~ ```python ~~AdjustLrFn: TypeAlias = Callable[[float, torch.Size], float]~~ ~~MsignFn: TypeAlias = Callable[[Tensor, BaseMsignFnConfig], Tensor]~~ ``` As discussed with team and in comment, we prefer to make the interface simpler and cleaner, thus we removed the callback interface, and canonicalize the original NS algorithm for Muon. The only configs available to users are `ns_steps`, `coefficients`, and `eps`, configurable through kwargs. By default, we use 5-step Newton-Schulz, with coefficients proposed by [Keller](https://kellerjordan.github.io/posts/muon/). We use LR adjustment proposed by [Moonshot](https://github.com/MoonshotAI/Moonlight/blob/master/Moonlight.pdf), which grafts learning rate from AdamW. **Testing** ~~1. Unit tests: the newly introduced Muon is covered in `test/test_optim.py`. We updated the test cases to pass named parameters to the optimizer under test. Additionally, we introduced a new test case to verify that when the user provides an empty FQN list, Muon correctly falls back to AdamW behavior.~~ As discussed, in order not to complicate the codebase, we prefer not to include reference implementation into PyTorch. We also updated the interface so we don't need to test the FQN based filtering. Muon is covered by the existing `test_optim.py` unit test. 2. End-to-end test: we added a training script that pre-trains a QWEN-like model on `openwebtext-100k` dataset. We trained for one epoch and the resulting loss curve is compared against the Moonshot implementation to confirm behavioral consistency. <img width="1102" height="472" alt="Screenshot 2025-07-29 at 1 04 12 AM" src="https://github.com/user-attachments/assets/ceab0733-497d-4070-8032-02ae7995c64c" /> **Numerics** We evaluate our implementation with existing implementation to confirm numerical consistency. As discussed, our implementation closely follows the algorithm described in [Keller's post](https://kellerjordan.github.io/posts/muon/), while incorporating the learning rate adjustment from [Moonlight](https://github.com/MoonshotAI/Moonlight/blob/master/Moonlight.pdf). This captures a key insight that allows users to reuse hyper-parameters tuned for `adamW`, making Muon a drop-in swap. As expected, the numerics difference mainly comes from `adjust_lr`, a max of ~5% relative diff in an example unit test setup below. ```python # dummy model and data model0 = Linear(10, 10, bias=False) model1 = copy.deepcopy(model0) inputs = torch.randn(8, 10) targets = torch.randn(8, 10) loss = MSELoss() lr = 1e-3 wd = 0.1 momentum = 0.95 opt_ref_muon = KellySingleDeviceMuon( params=model0.parameters(), lr=lr, weight_decay=wd, momentum=momentum, ) opt_exp_muon = Muon( params=model1.parameters(), lr=lr, weight_decay=wd, momentum=momentum, ) out_ref = model0(inputs) loss_ref = loss(out_ref, targets) opt_ref_muon.zero_grad() loss_ref.backward() opt_ref_muon.step() out_exp = model1(inputs) loss_exp = loss(out_exp, targets) opt_exp_muon.zero_grad() loss_exp.backward() opt_exp_muon.step() for p_ref, p_exp in zip(model0.parameters(), model1.parameters()): torch.testing.assert_close(p_ref, p_exp) ``` As explained above, including this `adjust_lr` is preferable. This is validated by an e2e training runs on training a qwen-2-like 0.5b model, where the curves show that training with `adjust_lr` converges more effectively than without. <img width="1179" height="464" alt="Screenshot 2025-08-18 at 10 12 33 AM" src="https://github.com/user-attachments/assets/e797d3da-c2f0-4187-b99e-5d48b7437c3c" /> **Performance** Training for one epoch of openwebtext-100k on eight H100 GPUs with DDP: - adamw_ddp finishes in 13.12 min - pytorch_muon_ddp finishes in 13.45 min Muon runs ~20s slower compared to AdamW. Assuming no other changes, Muon is *2.5%* slower than AdamW. AdamW: Optimizer.step() takes ~13.5 ms, step time ~930 ms <img width="726" height="590" alt="Screenshot 2025-07-29 at 1 56 14 AM" src="https://github.com/user-attachments/assets/ebcd7e1c-d129-4b20-9396-39f568edf03d" /> Muon: Optimizer.step() takes ~54 ms, step time ~960 ms <img width="751" height="597" alt="Screenshot 2025-07-29 at 2 02 20 AM" src="https://github.com/user-attachments/assets/72f5b904-ebd5-4502-a6ff-d3e9e5a6da81" /> **Note** We restrict the implementation to accept only 2D parameters. An alternative approach is to allow parameters with more than two dimensions and apply orthogonalization over the last two dimensions. We opt not to go with this approach as it can be error-prone. For example, with a kernel shaped `[in_channel, height, width, out_channel]`, applying orthogonalization to the last two dimensions is not meaningful. Since Muon is designed to operate orthogonalization on 2D matrices, preserving this assumption keeps the implementation clean and sound. **Next Steps** 1. Add `MuP` 2. Open-source optimized triton kernel for symmetric matmul. A preliminary benchmark found 1.23x - 1.48x speedup on small - large (n = 256 -> 16384) matrices. 3. Open-source unsharded Muon co-designed with FSDP2. **** Pull Request resolved: https://github.com/pytorch/pytorch/pull/160213 Approved by: https://github.com/janeyx99
67 lines
2.2 KiB
Python
67 lines
2.2 KiB
Python
"""
|
|
:mod:`torch.optim` is a package implementing various optimization algorithms.
|
|
|
|
Most commonly used methods are already supported, and the interface is general
|
|
enough, so that more sophisticated ones can also be easily integrated in the
|
|
future.
|
|
"""
|
|
|
|
from torch.optim import lr_scheduler as lr_scheduler, swa_utils as swa_utils
|
|
from torch.optim._adafactor import Adafactor as Adafactor
|
|
from torch.optim._muon import Muon as Muon
|
|
from torch.optim.adadelta import Adadelta as Adadelta
|
|
from torch.optim.adagrad import Adagrad as Adagrad
|
|
from torch.optim.adam import Adam as Adam
|
|
from torch.optim.adamax import Adamax as Adamax
|
|
from torch.optim.adamw import AdamW as AdamW
|
|
from torch.optim.asgd import ASGD as ASGD
|
|
from torch.optim.lbfgs import LBFGS as LBFGS
|
|
from torch.optim.nadam import NAdam as NAdam
|
|
from torch.optim.optimizer import Optimizer as Optimizer
|
|
from torch.optim.radam import RAdam as RAdam
|
|
from torch.optim.rmsprop import RMSprop as RMSprop
|
|
from torch.optim.rprop import Rprop as Rprop
|
|
from torch.optim.sgd import SGD as SGD
|
|
from torch.optim.sparse_adam import SparseAdam as SparseAdam
|
|
|
|
|
|
Adafactor.__module__ = "torch.optim"
|
|
Muon.__module__ = "torch.optim"
|
|
|
|
|
|
del adadelta # type: ignore[name-defined] # noqa: F821
|
|
del adagrad # type: ignore[name-defined] # noqa: F821
|
|
del adam # type: ignore[name-defined] # noqa: F821
|
|
del adamw # type: ignore[name-defined] # noqa: F821
|
|
del sparse_adam # type: ignore[name-defined] # noqa: F821
|
|
del adamax # type: ignore[name-defined] # noqa: F821
|
|
del asgd # type: ignore[name-defined] # noqa: F821
|
|
del sgd # type: ignore[name-defined] # noqa: F821
|
|
del radam # type: ignore[name-defined] # noqa: F821
|
|
del rprop # type: ignore[name-defined] # noqa: F821
|
|
del rmsprop # type: ignore[name-defined] # noqa: F821
|
|
del optimizer # type: ignore[name-defined] # noqa: F821
|
|
del nadam # type: ignore[name-defined] # noqa: F821
|
|
del lbfgs # type: ignore[name-defined] # noqa: F821
|
|
|
|
__all__ = [
|
|
"Adafactor",
|
|
"Adadelta",
|
|
"Adagrad",
|
|
"Adam",
|
|
"Adamax",
|
|
"AdamW",
|
|
"ASGD",
|
|
"LBFGS",
|
|
"lr_scheduler",
|
|
"Muon",
|
|
"NAdam",
|
|
"Optimizer",
|
|
"RAdam",
|
|
"RMSprop",
|
|
"Rprop",
|
|
"SGD",
|
|
"SparseAdam",
|
|
"swa_utils",
|
|
]
|