torchzero 0.1.3__py3-none-any.whl → 0.1.5__py3-none-any.whl

torchzero/modules/scheduling/step_size.py

@@ -30,24 +30,24 @@ class PolyakStepSize(OptimizerModule):
         self.use_grad = use_grad
         self.parameterwise = parameterwise
 
-    def _update(self, state, ascent):
-        if state.closure is None: raise ValueError("PolyakStepSize requires closure")
-        if state.fx0 is None: state.fx0 = state.closure(False) # can only happen when placed after SPSA
+    def _update(self, vars, ascent):
+        if vars.closure is None: raise ValueError("PolyakStepSize requires closure")
+        if vars.fx0 is None: vars.fx0 = vars.closure(False) # can only happen when placed after SPSA
 
         alpha = self.get_group_key('alpha')
 
         if self.parameterwise:
-            if self.use_grad: denom = (ascent*state.maybe_compute_grad_(self.get_params())).mean()
+            if self.use_grad: denom = (ascent*vars.maybe_compute_grad_(self.get_params())).mean()
             else: denom = ascent.pow(2).mean()
-            polyak_step_size: TensorList | Any = (state.fx0 - self.min_obj_value) / denom.where(denom!=0, 1) # type:ignore
+            polyak_step_size: TensorList | Any = (vars.fx0 - self.min_obj_value) / denom.where(denom!=0, 1) # type:ignore
             polyak_step_size = polyak_step_size.where(denom != 0, 0)
             if self.max is not None: polyak_step_size = polyak_step_size.clamp_max(self.max)
 
         else:
-            if self.use_grad: denom = (ascent*state.maybe_compute_grad_(self.get_params())).total_mean()
+            if self.use_grad: denom = (ascent*vars.maybe_compute_grad_(self.get_params())).total_mean()
             else: denom = ascent.pow(2).total_mean()
             if denom == 0: polyak_step_size = 0 # we converged
-            else: polyak_step_size = (state.fx0 - self.min_obj_value) / denom
+            else: polyak_step_size = (vars.fx0 - self.min_obj_value) / denom
 
             if self.max is not None:
                 if polyak_step_size > self.max: polyak_step_size = self.max
@@ -72,7 +72,7 @@ class RandomStepSize(OptimizerModule):
         self.low = low; self.high = high
         self.parameterwise = parameterwise
 
-    def _update(self, state, ascent):
+    def _update(self, vars, ascent):
         if self.parameterwise:
             lr = [random.uniform(self.low, self.high) for _ in range(len(ascent))]
         else:

torchzero/modules/second_order/newton.py

@@ -121,16 +121,16 @@ class ExactNewton(OptimizerModule):
         self.diag = diag
 
     @torch.no_grad
-    def step(self, state):
-        if state.closure is None: raise ValueError("Newton requires a closure to compute the gradient.")
+    def step(self, vars):
+        if vars.closure is None: raise ValueError("Newton requires a closure to compute the gradient.")
 
         params = self.get_params()
 
         # exact hessian via autograd
         with torch.enable_grad():
-            state.fx0 = state.closure(False)
-            grads, hessian = jacobian_and_hessian([state.fx0], params) # type:ignore
-            state.grad = grads = TensorList(grads).squeeze_(0)
+            vars.fx0 = vars.closure(False)
+            grads, hessian = jacobian_and_hessian([vars.fx0], params) # type:ignore
+            vars.grad = grads = TensorList(grads).squeeze_(0)
             gvec = grads.to_vec()
             hessian = hessian_list_to_mat(hessian)
 
@@ -148,18 +148,18 @@ class ExactNewton(OptimizerModule):
                     newton_step, success = _fallback_gd(hessian, gvec)
 
         # apply the `_update` method
-        state.ascent = grads.from_vec(newton_step.squeeze_().nan_to_num_(0,0,0))
+        vars.ascent = grads.from_vec(newton_step.squeeze_().nan_to_num_(0,0,0))
 
         # validate if newton step decreased loss
         if self.validate:
 
-            params.sub_(state.ascent)
-            fx1 = state.closure(False)
-            params.add_(state.ascent)
+            params.sub_(vars.ascent)
+            fx1 = vars.closure(False)
+            params.add_(vars.ascent)
 
             # if loss increases, set ascent direction to grad times lr
-            if (not fx1.isfinite()) or fx1 - state.fx0 > state.fx0 * self.tol: # type:ignore
-                state.ascent = grads.div_(grads.total_vector_norm(2) / self.gd_lr)
+            if (not fx1.isfinite()) or fx1 - vars.fx0 > vars.fx0 * self.tol: # type:ignore
+                vars.ascent = grads.div_(grads.total_vector_norm(2) / self.gd_lr)
 
         # peform an update with the ascent direction, or pass it to the child.
-        return self._update_params_or_step_with_next(state, params=params)
+        return self._update_params_or_step_with_next(vars, params=params)

torchzero/modules/smoothing/__init__.py

@@ -2,4 +2,4 @@ r"""
 Gradient smoothing and orthogonalization methods.
 """
 from .laplacian_smoothing import LaplacianSmoothing, gradient_laplacian_smoothing_
-from .gaussian_smoothing import GaussianSmoothing
+from .gaussian_smoothing import GaussianHomotopy

torchzero/modules/smoothing/gaussian_smoothing.py

@@ -5,7 +5,7 @@ import torch
 
 from ...tensorlist import TensorList, Distributions, mean as tlmean
 from ...utils.python_tools import _ScalarLoss
-from ...core import _ClosureType, OptimizationState, OptimizerModule, _maybe_pass_backward
+from ...core import _ClosureType, OptimizationVars, OptimizerModule, _maybe_pass_backward
 
 
 def _numpy_or_torch_mean(losses: list):
@@ -14,7 +14,7 @@ def _numpy_or_torch_mean(losses: list):
         return torch.mean(torch.stack(losses))
     return np.mean(losses).item()
 
-class GaussianSmoothing(OptimizerModule):
+class GaussianHomotopy(OptimizerModule):
     """Samples and averages value and gradients in multiple random points around current position.
     This effectively applies smoothing to the function.
 
@@ -43,9 +43,9 @@ class GaussianSmoothing(OptimizerModule):
         self.sample_x0 = sample_x0
 
     @torch.no_grad()
-    def step(self, state: OptimizationState):
-        if state.closure is None: raise ValueError('GaussianSmoothing requires closure.')
-        closure = state.closure
+    def step(self, vars: OptimizationVars):
+        if vars.closure is None: raise ValueError('GaussianSmoothing requires closure.')
+        closure = vars.closure
         params = self.get_params()
         sigmas = self.get_group_key('sigma')
 
@@ -83,8 +83,8 @@ class GaussianSmoothing(OptimizerModule):
 
 
         self.current_step += 1
-        state.closure = smooth_closure
-        return self._update_params_or_step_with_next(state)
+        vars.closure = smooth_closure
+        return self._update_params_or_step_with_next(vars)
 
 
 # todo single loop gaussian homotopy?

torchzero/modules/smoothing/laplacian_smoothing.py

@@ -94,7 +94,7 @@ class LaplacianSmoothing(OptimizerModule):
 
 
     @torch.no_grad
-    def _update(self, state, ascent):
+    def _update(self, vars, ascent):
         params = self.get_params()
         sigmas = self.get_group_key('sigma')
 

torchzero/modules/weight_averaging/ema.py

@@ -50,7 +50,7 @@ class SwitchEMA(OptimizerModule):
         params.set_(self.orig_params)
 
     @torch.no_grad
-    def step(self, state):
+    def step(self, vars):
         # if self.next_module is not None:
         #     warn(f'EMA should usually be the last module, but {self.next_module.__class__.__name__} is after it.')
         self.cur_step += 1
@@ -58,7 +58,7 @@ class SwitchEMA(OptimizerModule):
         params = self.get_params()
         # state.maybe_use_grad_(params)
         # update params with the child. Averaging is always applied at the end.
-        ret = self._update_params_or_step_with_next(state, params)
+        ret = self._update_params_or_step_with_next(vars, params)
 
         ema = self.get_state_key('ema', init = 'params', params=params)
         momentum = self.get_group_key('momentum')
@@ -67,6 +67,6 @@ class SwitchEMA(OptimizerModule):
 
         if (self.update_every is not None) and (self.cur_step % self.update_every == 0):
             params.set_(ema.clone())
-            if self._reset_stats: state.add_post_step_hook(_reset_stats_hook)
+            if self._reset_stats: vars.add_post_step_hook(_reset_stats_hook)
 
         return ret

torchzero/modules/weight_averaging/swa.py

@@ -55,10 +55,10 @@ class PeriodicSWA(OptimizerModule):
         self.swa_cur = 0
         self.n_models = 0
 
-    def step(self, state):
+    def step(self, vars):
         swa = None
         params = self.get_params()
-        ret = self._update_params_or_step_with_next(state, params)
+        ret = self._update_params_or_step_with_next(vars, params)
 
         # start first period after `pswa_start` steps
         if self.cur >= self.pswa_start:
@@ -88,7 +88,7 @@ class PeriodicSWA(OptimizerModule):
 
             params.set_(swa)
             # add a hook that resets momentum, which also deletes `swa` in this module
-            if self._reset_stats: state.add_post_step_hook(_reset_stats_hook)
+            if self._reset_stats: vars.add_post_step_hook(_reset_stats_hook)
 
         return ret
 
@@ -127,13 +127,13 @@ class CyclicSWA(OptimizerModule):
 
         self.cur_lr = self.init_lr
 
-    def step(self, state):
+    def step(self, vars):
         params = self.get_params()
 
         # start first period after `cswa_start` steps
         if self.cur >= self.cswa_start:
 
-            ascent = state.maybe_use_grad_(params)
+            ascent = vars.maybe_use_grad_(params)
 
             # determine the lr
             point = self.cycle_cur / self.cycle_length
@@ -146,7 +146,7 @@ class CyclicSWA(OptimizerModule):
                 lr = init_lr * (1-p2) + peak_lr * p2
 
             ascent *= lr
-            ret = self._update_params_or_step_with_next(state, params)
+            ret = self._update_params_or_step_with_next(vars, params)
 
             if self.sample_all or self.cycle_cur in (0, self.cycle_length, self.cycle_length // 2):
                 swa = self.get_state_key('swa')
@@ -159,12 +159,12 @@ class CyclicSWA(OptimizerModule):
                     self.cycle_cur = -1
 
                     params.set_(swa)
-                    if self._reset_stats: state.add_post_step_hook(_reset_stats_hook)
+                    if self._reset_stats: vars.add_post_step_hook(_reset_stats_hook)
 
             self.cycle_cur += 1
 
         else:
-            ret = self._update_params_or_step_with_next(state, params)
+            ret = self._update_params_or_step_with_next(vars, params)
 
         self.cur += 1
 

torchzero/optim/first_order/forward_gradient.py

@@ -2,7 +2,7 @@ from typing import Literal
 
 import torch
 
-from ...core import OptimizationState, OptimizerModule
+from ...core import OptimizationVars, OptimizerModule
 from ...modules import ForwardGradient as _ForwardGradient, SGD, WeightDecay, LR
 from ...tensorlist import Distributions
 from ..modular import Modular

torchzero/optim/modular.py

@@ -3,7 +3,7 @@ import warnings
 from inspect import cleandoc
 import torch
 
-from ..core import OptimizerModule, TensorListOptimizer, OptimizationState, _Chain, _Chainable
+from ..core import OptimizerModule, TensorListOptimizer, OptimizationVars, _Chain, _Chainable
 from ..utils.python_tools import flatten
 
 def _unroll_modules(flat_modules: list[OptimizerModule], nested) -> list[OptimizerModule]:
@@ -126,7 +126,7 @@ class Modular(TensorListOptimizer):
         raise ValueError(f'No modules out of {", ".join(m.__class__.__name__ for m in modules)} match "{name}".')
 
     def step(self, closure=None): # type:ignore
-        state = OptimizationState(closure, self.model)
-        res = self.chain.step(state)
-        for hook in state.post_step_hooks: hook(self, state)
+        vars = OptimizationVars(closure, self.model)
+        res = self.chain.step(vars)
+        for hook in vars.post_step_hooks: hook(self, vars)
         return res

torchzero/tensorlist.py

@@ -10,7 +10,7 @@ in an optimizer when you have to create one from parameters on each step. The so
 it once beforehand, but then you won't be able to easily support parameter groups and per-parameter states.
 """
 import builtins
-from collections.abc import Callable, Sequence, Iterable, Generator
+from collections.abc import Callable, Sequence, Iterable, Generator, Iterator
 import math
 import operator
 from typing import Any, Literal, TypedDict
@@ -744,6 +744,13 @@ class NumberList(TensorList):
     Note that this only supports basic arithmetic operations that are overloaded.
 
     Can't use a numpy array because _foreach methods do not work with it."""
+    # remove torch.Tensor from return values
+    def __getitem__(self, i) -> Any:
+        return super().__getitem__(i)
+
+    def __iter__(self) -> Iterator[Any]:
+        return super().__iter__()
+
     def _set_to_method_result_(self, method: str, *args, **kwargs):
         """Sets each element of the tensorlist to the result of calling the specified method on the corresponding element.
         This is used to support/mimic in-place operations."""

{torchzero-0.1.3.dist-info → torchzero-0.1.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: torchzero
-Version: 0.1.3
+Version: 0.1.5
 Summary: Modular optimization library for PyTorch.
 Author-email: Ivan Nikishev <nkshv2@gmail.com>
 License: MIT License

torchzero-0.1.5.dist-info/RECORD

@@ -0,0 +1,104 @@
+torchzero/__init__.py,sha256=CCIYfhGNYMnRP_cdXL7DgocxkEWYUZYgB3Sf1T5tdYY,203
+torchzero/tensorlist.py,sha256=V9m5zJ44PtiJxTOiM7cADxYaWK-ogv5Tk_KnyQqm1oo,41601
+torchzero/core/__init__.py,sha256=hab7HAep0JIVeJ-EQhcOAB9oKIIo2MuCVn7yS3BFVYA,266
+torchzero/core/module.py,sha256=Dhf8Rn6zKpzzO3rzyxzJnRAe_tOKS35AgH-FQQKR16A,22362
+torchzero/core/tensorlist_optimizer.py,sha256=MenRzyNPQJWx26j4Tj2vbX2pEcRu_26HQPIpUu_nsFc,9957
+torchzero/modules/__init__.py,sha256=5f8kt2mMn1eo9YcjXc3ESW-bqMQIRf646V3zlr8UAO4,571
+torchzero/modules/adaptive/__init__.py,sha256=YBVDXCosr4-C-GFCoreHS3DFyHiYMhCbOWgdhVVaZ_E,161
+torchzero/modules/adaptive/adaptive.py,sha256=Zp6cBbuLIeyEDcZJxgpdXPVnw7sjhkBSe9Pd3HHkHXk,6477
+torchzero/modules/experimental/__init__.py,sha256=dRLPbhTO8efkRQHB-8Z29dKS3BEleCso1dsXxgccFgM,647
+torchzero/modules/experimental/experimental.py,sha256=4w3WyBL43WWwYL1ZdoCAdtBHnJfTrUYe9NBmmqTxYes,9808
+torchzero/modules/experimental/quad_interp.py,sha256=6Nsvpv580KpSujPYHxK-qB5m6b7C_dFdBxCXE49VHAQ,4106
+torchzero/modules/experimental/subspace.py,sha256=6n0INUqya24RR4Onm7bGNAPsdqw4lPYIGqT6OCwXNOU,11766
+torchzero/modules/gradient_approximation/__init__.py,sha256=q8rNkk2PK6Y9zk42Mq8pY2YF6fHt5QuJd7BK-RTFKpg,179
+torchzero/modules/gradient_approximation/_fd_formulas.py,sha256=mXqRwsDYjXi3pnI4mkpwwtJE85omYEvcmtbJAOfpg9o,82
+torchzero/modules/gradient_approximation/base_approximator.py,sha256=VjQ_NNLElDnk4kGPYy8iK9A6f2Y6yo6H1axpXkkAeSw,3632
+torchzero/modules/gradient_approximation/fdm.py,sha256=1FVPe3J3tOBgktCKlcd-wOv3bo_FynuF9U4RKbsz-CY,4976
+torchzero/modules/gradient_approximation/forward_gradient.py,sha256=3CC-tcUIAL0d422WHUQLPx0UMcU5URQ5uYCNbgpi19M,6456
+torchzero/modules/gradient_approximation/newton_fdm.py,sha256=BqkNZp00pUBJLWAXKNTk2eggbE4o6u-eLmvWw55zihg,7100
+torchzero/modules/gradient_approximation/rfdm.py,sha256=2nYl_Pu6Lc3gUjTu0GqTImoMSCX5MPFKmvoagpdrf0w,4966
+torchzero/modules/line_search/__init__.py,sha256=hYysFi5b9a5g0jcRNkgZYGRcZ1V7_JacBVWMR7idP38,1380
+torchzero/modules/line_search/armijo.py,sha256=Iw7bsx2yhDhZ0POvD2lc28tv_b5zphOq3MPed8dAYLk,1955
+torchzero/modules/line_search/base_ls.py,sha256=tPlJxZ_CVGPxVrG_jxud8LRbgceZKMf-ZBbDv0jJX2Q,5319
+torchzero/modules/line_search/directional_newton.py,sha256=PMNjjj6n1GVuMe9fwGGYnxbA9Tw5CQxLXgJNhNiwURs,9148
+torchzero/modules/line_search/grid_ls.py,sha256=tCrDxE1OMIXMisYdyiBF56XBRP16mbSs5SFDRcy1vCw,5603
+torchzero/modules/line_search/scipy_minimize_scalar.py,sha256=STGvMuCPcqxw9iaiD6jLsI6YgxAvEic0tfqDQ90qAiA,2196
+torchzero/modules/meta/__init__.py,sha256=ARVR3Vzvq50n-3uFMNxcGUDd2r4Euamay5UYtpIxXNg,407
+torchzero/modules/meta/alternate.py,sha256=kJB3Y507u_a_dra7N0Wo4uTQhbMSfizG0mmmuISzeH8,2119
+torchzero/modules/meta/grafting.py,sha256=IenyBO0FmJDO6st_mfFAUnu98UlChm4Io4aKARsDI0s,7560
+torchzero/modules/meta/optimizer_wrapper.py,sha256=x1nkKoqGK6-lgiuRYzVFcTNUCYeiwFet0MZoaQ3kv8U,6432
+torchzero/modules/meta/return_overrides.py,sha256=Amp4_yiMtXNM8_JgnAKK71BP-NrqvkYha4wt8mA-Nb4,2023
+torchzero/modules/misc/__init__.py,sha256=P43XRz1nnOuJbpq_bQboLJ7hip80SQmvhua2baPdJ-c,390
+torchzero/modules/misc/accumulate.py,sha256=M66TaL0XBioC-2WkPOj8crKoQrWx-np-Im4I2pzvwHU,1324
+torchzero/modules/misc/basic.py,sha256=EnuS6nZwyRXjWytyQef3f5885NC2b79wiQeZSU-urYA,3475
+torchzero/modules/misc/lr.py,sha256=xG1TCPeZazp1SBkAdwxY3tbt9-Tgufzq0noPlcsZkO4,3523
+torchzero/modules/misc/multistep.py,sha256=Jnbs1upMpvabhSkju10ciOuPzMd0CD5UoK4ueUrh048,1770
+torchzero/modules/misc/on_increase.py,sha256=WT86io0qtTk4UbrNAQlXyPD7ad61EagKqhr2mrrRcYA,1998
+torchzero/modules/momentum/__init__.py,sha256=Cj_3KJ76RLX-WQ7xsOoLY9mucQvnkyudTeVH8fnvdwc,138
+torchzero/modules/momentum/momentum.py,sha256=3zpmhzOI7KLX2nbaWivGEquo9nboEmkIyAO8k8SmWdw,3931
+torchzero/modules/operations/__init__.py,sha256=4SxIQMh-ixEqEDXWdizeGOzfhFw-af4Wz0A-00ypmg0,378
+torchzero/modules/operations/multi.py,sha256=3T0mqrnKrdE1G7wayolCZVjWWjLG7geafu1vbClmz3Y,10273
+torchzero/modules/operations/reduction.py,sha256=mMqxW-PO7T-azTnbx2vFnfOVTHfw7u7BzZveoaaCfxc,4621
+torchzero/modules/operations/singular.py,sha256=J1_0HKPi6Tu0j-25lMP4FPRwgjm8drM5QRcUCmvQNV8,3560
+torchzero/modules/optimizers/__init__.py,sha256=QZu8yvqy7ouY45Xju5Z2oHWJiFa1CslknodhWWRZRms,247
+torchzero/modules/optimizers/adagrad.py,sha256=7SQiQBtkN72nhZ6a22p4KCCwzxUYcvt1qPRgT_y2I7w,2014
+torchzero/modules/optimizers/adam.py,sha256=17HJl5VHtUDHPVSW8nTVb9XrjuXQ1H6lFLAaGUkq028,4714
+torchzero/modules/optimizers/lion.py,sha256=DRm09OpQSs2YjFx21OMA5ADp6dVYLPN6uu61GkxyF6o,904
+torchzero/modules/optimizers/rmsprop.py,sha256=vcPjGh1krMLGX97sNKgKc5rx0-tT_cQznSwgmEtlq0w,2152
+torchzero/modules/optimizers/rprop.py,sha256=OjLmHhhPjfRe0KxdPSKduIQs5atwoCvbNLLoDbFWAfI,3603
+torchzero/modules/optimizers/sgd.py,sha256=t31zfKuV6tJRzIOEtZBX0gH_KGdRi4ddvKhYibK5eCM,2242
+torchzero/modules/orthogonalization/__init__.py,sha256=brvrj57U_1qKKU8AUqbe9lyY9jsfzZvUGnvsU4wjDSQ,151
+torchzero/modules/orthogonalization/newtonschulz.py,sha256=gXSkONee9GC5oTODWde2d2HaxTVWbSZMIG1ZD9p5LRE,7517
+torchzero/modules/orthogonalization/svd.py,sha256=HO72YUP4Jlk-WeCoVpT3kqPLuiUwcApNvsXx3-PMFAI,3651
+torchzero/modules/quasi_newton/__init__.py,sha256=G5lW-q0pI47SJ2AZuY4qkjbqfYzJS0qATDo7V0BGzD4,124
+torchzero/modules/regularization/__init__.py,sha256=FD_KERcYY4bdVR22OuKXUUVt63jyfE9V84evwDC1edo,498
+torchzero/modules/regularization/dropout.py,sha256=Wt0rHQxMt2t28fNaNkfNlXVcbcf3TlqWhKzWi1kz6lU,1003
+torchzero/modules/regularization/noise.py,sha256=LFSG4Ga5W2VuT1jzkApKodMiDsHbi8zYJ9N_ZnsLMoI,3026
+torchzero/modules/regularization/normalization.py,sha256=LxegX6Wf0QYZPSxjtUsx7x6IO9UCGc_SBYRfFp3EJHk,12152
+torchzero/modules/regularization/ortho_grad.py,sha256=TLaF8GIM6_tbHHvUhi9XUKJvTdPgF9RM364HXYmVliw,3126
+torchzero/modules/regularization/weight_decay.py,sha256=AjR1RuUDa2DM1PbHx-wBT41XaaAs2w5PKLjRRpV3JrU,3624
+torchzero/modules/scheduling/__init__.py,sha256=NxR1cpKXtZSbVqPRlzzzgH3_JBMuxQCf3nUhmxBN2Cc,89
+torchzero/modules/scheduling/lr_schedulers.py,sha256=MYcQ13YexvyTSD9AHxdu54DSu_neTxY-BImUwBFc7Iw,4963
+torchzero/modules/scheduling/step_size.py,sha256=NTHOFO34GroEE_O2IWmZprRpD9wiyLu7iNvaAUlfVbc,3664
+torchzero/modules/second_order/__init__.py,sha256=oRyRy8mCjurMINHNdsxjlptLbQNU0VnstkDm1Ccv_80,182
+torchzero/modules/second_order/newton.py,sha256=iVNawkyiz-Yj8_cAKi1BYeffk9P0G9UoddnVTBogR64,6751
+torchzero/modules/smoothing/__init__.py,sha256=sKc9P7QJUNEyuoQXjhMkAELFdqe11tB1HnnKeMI4ciQ,194
+torchzero/modules/smoothing/gaussian_smoothing.py,sha256=kP72r-f0prGEEzgaVeY1RLFLTfx05L_SOtLMCrkXeOs,3827
+torchzero/modules/smoothing/laplacian_smoothing.py,sha256=ehb0lj479RRKZTJ5_9FGDXwidWF5soo9hw8vquqrl5s,5552
+torchzero/modules/weight_averaging/__init__.py,sha256=nJJRs68AV2G4rGwiiHNRfm6XmtM-xUev1pCtzNIVfa8,66
+torchzero/modules/weight_averaging/ema.py,sha256=xdVnKC8PxCQCec4ad3ncvznvVsQM5O6EVfsOKsRr18k,2854
+torchzero/modules/weight_averaging/swa.py,sha256=syio5qq1vf76d5bAJU-I8Zc8U12bR2n8C9hEznCgr7s,6764
+torchzero/optim/__init__.py,sha256=vk6pIYJHWAGYJMdtJ1otsmVph-pdL5HwBg-CTeBCGso,253
+torchzero/optim/modular.py,sha256=tiuKXpLqq6jWwWWrfC_QU9XkErl2Saaw_xGau0frwzQ,6060
+torchzero/optim/experimental/__init__.py,sha256=RqNzJu5mVl3T0u7cf4TBzSiA20M1kxTZVYWjSVhEHuU,585
+torchzero/optim/experimental/experimental.py,sha256=tMHZVbEXm3s6mMr7unFSvk_Jks3uAaAG0fzsH6gr098,10928
+torchzero/optim/experimental/ray_search.py,sha256=GYyssL64D6RiImrZ2tchoZJ04x9rX-Bp1y2nQXEGxX0,2662
+torchzero/optim/first_order/__init__.py,sha256=CRT4farcwi8sO1qqDGxXv1856zOwuKlJKBIAIvpL2Z0,336
+torchzero/optim/first_order/cautious.py,sha256=XBeqrLQ4gFKVUYnJI5ROmF9wQJGhY90HR6UG5IS7vYk,6610
+torchzero/optim/first_order/forward_gradient.py,sha256=8u_ya0NuFSopPPhOiqqSov35irg8tF3LeHYg-qthTG0,3065
+torchzero/optim/first_order/optimizers.py,sha256=jYmU6YDsYRGMRsCNkYc6AlvOf3wlU7Uv1xUrzl0o8zo,24501
+torchzero/optim/quasi_newton/__init__.py,sha256=0X83dl-85_j3ck8itWxJR49ZbFeOcWurW6FI8J12F1w,49
+torchzero/optim/quasi_newton/directional_newton.py,sha256=oZ-If8SRcFXTDFKS_zlAcJnif-v5dTCR9HXqmfvsvNA,2595
+torchzero/optim/second_order/__init__.py,sha256=3Gt0dR4NzBK07TV0NF8KZImUGHbI8E2zncDmhIC377I,31
+torchzero/optim/second_order/newton.py,sha256=-DqJrS8JPea6Y9jp5lDV14KyP8A24YKiPxJ32Vsfiv4,3848
+torchzero/optim/wrappers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+torchzero/optim/wrappers/nevergrad.py,sha256=4PLqfs2L9XJhveyX6l7kJu1cIPl6uv7_UD76amIlP7I,4733
+torchzero/optim/wrappers/nlopt.py,sha256=fGDOZ82sRI2VLH3hKIAhZY4EuKSdu_g217c-NZvD_rs,7104
+torchzero/optim/wrappers/scipy.py,sha256=_BQwFDQ7SBqIA5i1SJ29Xj0jDXVV8MQ_9RcsPT3U6VQ,18047
+torchzero/optim/zeroth_order/__init__.py,sha256=_6T0znO6V63Niq7DMhJPgUuMc_nPvAGxjCjMdf-r64U,218
+torchzero/optim/zeroth_order/fdm.py,sha256=5iJc_F_tRR4cGQfy2Jr8PmAnCGrPva89ZWczSdcBkFk,3686
+torchzero/optim/zeroth_order/newton_fdm.py,sha256=-5E1FGzeJMr8_IougzE_FEOPFt9pEjQxID4Y89Hpmh0,6537
+torchzero/optim/zeroth_order/rfdm.py,sha256=_Y7yiF1bsVRlXt5IK-3zQccwVl95JF0-Xw-fl8Q_7y4,10529
+torchzero/optim/zeroth_order/rs.py,sha256=3w2nnPGWPecourEdUG583vchcqdNxC6Q_PBL3l0PvCk,3333
+torchzero/random/__init__.py,sha256=8EowQhC4yTZuF8w1ZDl73YZtLej8SuhxCk1Bkifbkms,93
+torchzero/random/random.py,sha256=Oq4GvM_6AOsabg5ke6b8h51V9npyHVxp8ob_wC5D-Aw,2829
+torchzero/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+torchzero/utils/compile.py,sha256=pYEyX8P26iCb_hFqAXC8IP2SSQrRfC7ZDhXS0vVCsfY,1257
+torchzero/utils/derivatives.py,sha256=koLmuUcVcX41SrH_9rvfJyMXyHyocNLuZ-C8Kr2B7hk,4844
+torchzero/utils/python_tools.py,sha256=kkyDhoP695HhapfKrdjcrRbRAbcvB0ArP1pkxuVUlf0,1192
+torchzero/utils/torch_tools.py,sha256=sSBY5Bmk9LOAgPtaq-6TK4wDgPXsg6FIWxv8CVDx82k,3580
+torchzero-0.1.5.dist-info/LICENSE,sha256=r9ZciAoZoqKC_FNADE0ORukj1p1XhLXEbegdsAyqhJs,1087
+torchzero-0.1.5.dist-info/METADATA,sha256=h7RXRQD3U-Gfa6gFJtE-ErJ_bYT8h0EdYsqkNg8kIUs,6096
+torchzero-0.1.5.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+torchzero-0.1.5.dist-info/top_level.txt,sha256=isztuDR1ZGo8p2tORLa-vNuomcbLj7Xd208lhd-pVPs,10
+torchzero-0.1.5.dist-info/RECORD,,

torchzero-0.1.3.dist-info/RECORD

@@ -1,104 +0,0 @@
-torchzero/__init__.py,sha256=CCIYfhGNYMnRP_cdXL7DgocxkEWYUZYgB3Sf1T5tdYY,203
-torchzero/tensorlist.py,sha256=5XvutWeZdyUcAVVkoEXbgSjvTiDe0f1806wsINfzlRg,41387
-torchzero/core/__init__.py,sha256=aw2p6Gt0qrtUM8x2msspAG8JHMaD2ma11sMhXm-rC90,267
-torchzero/core/module.py,sha256=msvh-e7cE3vArXb5PLrrsG3j7d-94GG3V4aBXYKFZTo,21254
-torchzero/core/tensorlist_optimizer.py,sha256=hIQIW7uBwddlJxXhR4-xnll0ixY9ZJY0i0oH-wqxfX0,9963
-torchzero/modules/__init__.py,sha256=5f8kt2mMn1eo9YcjXc3ESW-bqMQIRf646V3zlr8UAO4,571
-torchzero/modules/adaptive/__init__.py,sha256=YBVDXCosr4-C-GFCoreHS3DFyHiYMhCbOWgdhVVaZ_E,161
-torchzero/modules/adaptive/adaptive.py,sha256=msQkLlxqt3fFWMLGyr1Pi77bGPv-QAyfDfQ1Oipl5Yo,6473
-torchzero/modules/experimental/__init__.py,sha256=dRLPbhTO8efkRQHB-8Z29dKS3BEleCso1dsXxgccFgM,647
-torchzero/modules/experimental/experimental.py,sha256=Z7g3AhZ5udtjIF_3a5-GdvtMOSt_uMmyJF8AKshw9i0,9853
-torchzero/modules/experimental/quad_interp.py,sha256=iAP6r2uHp2BDsGNFYhpIHuv1l5z7ZXQZLZU1E-VO7eE,4117
-torchzero/modules/experimental/subspace.py,sha256=KgTIdzv75hlmPFs9nCMV0M3NWa7zcp9IK0OLFsl5UrU,11801
-torchzero/modules/gradient_approximation/__init__.py,sha256=q8rNkk2PK6Y9zk42Mq8pY2YF6fHt5QuJd7BK-RTFKpg,179
-torchzero/modules/gradient_approximation/_fd_formulas.py,sha256=mXqRwsDYjXi3pnI4mkpwwtJE85omYEvcmtbJAOfpg9o,82
-torchzero/modules/gradient_approximation/base_approximator.py,sha256=lNDrbGrsfhru7u40w2EDd5Xccv9oQxRaEDOLoVnBJdk,3975
-torchzero/modules/gradient_approximation/fdm.py,sha256=yBvWmkKJqVMd0TUSZcEr6_ulNTtlZbE7rl0uLI7BADY,4977
-torchzero/modules/gradient_approximation/forward_gradient.py,sha256=3CC-tcUIAL0d422WHUQLPx0UMcU5URQ5uYCNbgpi19M,6456
-torchzero/modules/gradient_approximation/newton_fdm.py,sha256=foEjnOUn-9Til6IP6x6E7-wVoAnKGMfbyk1hfmuRYcI,7116
-torchzero/modules/gradient_approximation/rfdm.py,sha256=iA2ydwkk9iRVPYcaECqUupprj6nrgmRKPVN0936gYf4,4967
-torchzero/modules/line_search/__init__.py,sha256=hYysFi5b9a5g0jcRNkgZYGRcZ1V7_JacBVWMR7idP38,1380
-torchzero/modules/line_search/armijo.py,sha256=2-tErIpO8p3XhY9uKrwGsaohEAN6h5tZ847_hXTPjxs,1966
-torchzero/modules/line_search/base_ls.py,sha256=uRHg6n9lcLrBrIAUZLuEiJuWaZDQR-rqNO0ZxZYGAXo,5330
-torchzero/modules/line_search/directional_newton.py,sha256=LLMjDu9nzy-WiqXb2Kzc4kRzhCoQOFc-1j_9hOyxt00,9168
-torchzero/modules/line_search/grid_ls.py,sha256=PLpi8R_KIc8xZ6IxJmeLgKPJQPSgd5M4T-pj33ykLnw,5614
-torchzero/modules/line_search/scipy_minimize_scalar.py,sha256=6JS603_sphNxj4Ji2Ia4gWcyqIM326MVGMHLWaQDXBA,2201
-torchzero/modules/meta/__init__.py,sha256=ARVR3Vzvq50n-3uFMNxcGUDd2r4Euamay5UYtpIxXNg,407
-torchzero/modules/meta/alternate.py,sha256=6LOYJI6_Q2X5MKAnFcymoJP8i4lcarhaPRp9Sm0eQS0,2124
-torchzero/modules/meta/grafting.py,sha256=tWazkxlqw2brkJjkaugasoh3XGaAsR4xOqyjQMEv5uQ,7583
-torchzero/modules/meta/optimizer_wrapper.py,sha256=cVe63uXMLLIbuj4wkSQyj0mBmKVx9SqSVzqsgcczbzA,6448
-torchzero/modules/meta/return_overrides.py,sha256=6bveA6P0jgNiWu-P2NumAjfrAtpOL_uoIHBljOu-aYs,2031
-torchzero/modules/misc/__init__.py,sha256=P43XRz1nnOuJbpq_bQboLJ7hip80SQmvhua2baPdJ-c,390
-torchzero/modules/misc/accumulate.py,sha256=qN0xJ-wnhH6pyr8OY5g0N-ObYJrPyiYgVjX11Sss10s,1330
-torchzero/modules/misc/basic.py,sha256=dWLMkj32bp4FN44wFrF5VP1_fyN7Xxb9FTVtWP_4EWw,3492
-torchzero/modules/misc/lr.py,sha256=V6W5AU9upjpSVpFtlXVar-zJ-qZAgY2Bts-ibeKf4bk,3525
-torchzero/modules/misc/multistep.py,sha256=L526iSNWg8UbXdulRT09r4qcOm6jHXi4v3Ho8PjkCPQ,1781
-torchzero/modules/misc/on_increase.py,sha256=XoMzB6VWOIKpujL030fpwQcVyW_QSls-ipCwjoveMF0,2012
-torchzero/modules/momentum/__init__.py,sha256=Cj_3KJ76RLX-WQ7xsOoLY9mucQvnkyudTeVH8fnvdwc,138
-torchzero/modules/momentum/momentum.py,sha256=Tywb6g0PNY4gIfXRHxEIYxgH56qoAnAtE5MzPAJh7VU,3935
-torchzero/modules/operations/__init__.py,sha256=4SxIQMh-ixEqEDXWdizeGOzfhFw-af4Wz0A-00ypmg0,378
-torchzero/modules/operations/multi.py,sha256=XValEBe90w1uXY94kX_DItWvjchmOrAfpUa6hsi0sxk,10317
-torchzero/modules/operations/reduction.py,sha256=uIRrqG2V-NOvljZVrJegjfjcCSQ4pI1rN9bQnZt-EW4,4652
-torchzero/modules/operations/singular.py,sha256=Y6ImnsUB3CJPpe_7UkT1vq9diGWXf7XKpuA7Ev7Hq2g,3569
-torchzero/modules/optimizers/__init__.py,sha256=QZu8yvqy7ouY45Xju5Z2oHWJiFa1CslknodhWWRZRms,247
-torchzero/modules/optimizers/adagrad.py,sha256=20r1ghs67NfCED0Z0xPZflen1gLaG3tOLiTg0WEYsNU,2015
-torchzero/modules/optimizers/adam.py,sha256=6Lq69rsyE_UI54z8T0HRyHqo0nXLwjCv35u6BsnpVSg,4722
-torchzero/modules/optimizers/lion.py,sha256=LkXedRYK_IxJ1Xebn9dzOOMOnozM_OXVD20_wqOIB2w,905
-torchzero/modules/optimizers/rmsprop.py,sha256=pqeaGxec-IY1i4Io5_iMZaef8nOKVZPVblVdibaWy40,2153
-torchzero/modules/optimizers/rprop.py,sha256=nvv-PPvEpGJs0imJNr0BPRV0X_bMrEY2TaUeF986sa0,3604
-torchzero/modules/optimizers/sgd.py,sha256=dCSQ1UwiycAR8-nSTLnk0G8sieiNAgUyChZsCwgvEOY,2246
-torchzero/modules/orthogonalization/__init__.py,sha256=brvrj57U_1qKKU8AUqbe9lyY9jsfzZvUGnvsU4wjDSQ,151
-torchzero/modules/orthogonalization/newtonschulz.py,sha256=IEqA_Udi2E5AbG2K_XZEz0O8GbGuKjNmQ1GVdEZP0xs,7520
-torchzero/modules/orthogonalization/svd.py,sha256=SsuPFjODwM7uJzOxXW8LbLAbU-4hlLBmp1Fh5xWDCW4,3652
-torchzero/modules/quasi_newton/__init__.py,sha256=G5lW-q0pI47SJ2AZuY4qkjbqfYzJS0qATDo7V0BGzD4,124
-torchzero/modules/regularization/__init__.py,sha256=FD_KERcYY4bdVR22OuKXUUVt63jyfE9V84evwDC1edo,498
-torchzero/modules/regularization/dropout.py,sha256=YlJmmYOVaYIoeQQW2z8kXZfRyXntZfg4tX0m6_w6JDo,1004
-torchzero/modules/regularization/noise.py,sha256=Z_BrotV5QE5HY4E6DhIpoSjsqejaCNm_n393euTtA3o,3014
-torchzero/modules/regularization/normalization.py,sha256=FE51Ww-aDXTQSJr-qj2YxBHRCOjluZC7TqCmXF3Xagc,12142
-torchzero/modules/regularization/ortho_grad.py,sha256=DnUYXAc8VCMSOS5NoZSf1XrU9TStdyt8QpU8bhBzYqE,3127
-torchzero/modules/regularization/weight_decay.py,sha256=4QeTpTra52MLyTrgCSaeaB8JxN-l8gVDq4JIwNoL41k,3625
-torchzero/modules/scheduling/__init__.py,sha256=NxR1cpKXtZSbVqPRlzzzgH3_JBMuxQCf3nUhmxBN2Cc,89
-torchzero/modules/scheduling/lr_schedulers.py,sha256=8zEK_wtE9IqnhHtS3FYNdh_f83q8V90YqLa1zWVzEW4,4965
-torchzero/modules/scheduling/step_size.py,sha256=UOE2ZIcVTX7FHlO8BUqtMy31_jmOKQMpgkkc-WgLfZs,3674
-torchzero/modules/second_order/__init__.py,sha256=oRyRy8mCjurMINHNdsxjlptLbQNU0VnstkDm1Ccv_80,182
-torchzero/modules/second_order/newton.py,sha256=RPn0kHg6ZCAZLQLFW82eQAh7B1-U6d70xTb-CHbJLUs,6765
-torchzero/modules/smoothing/__init__.py,sha256=-mxey48zc72aGV0dv4TLHeFpf98QZjlxMu5Pct1LI_Y,195
-torchzero/modules/smoothing/gaussian_smoothing.py,sha256=9oxVMv--B0ESzOrhEaqQQeTWaVrSIqJXcU77VaRB2KE,3835
-torchzero/modules/smoothing/laplacian_smoothing.py,sha256=TXy2IgVqZehH97PQWn655mK7nDlNEr3EeeCkKEVT0tA,5553
-torchzero/modules/weight_averaging/__init__.py,sha256=nJJRs68AV2G4rGwiiHNRfm6XmtM-xUev1pCtzNIVfa8,66
-torchzero/modules/weight_averaging/ema.py,sha256=tun6TNOMQWeAZyvkbJEDLf3tGgvJPhhWAAA5ScBsT08,2857
-torchzero/modules/weight_averaging/swa.py,sha256=A4nRGQyMnZ2CaOW20iVfAs_iqV3lnULt7t--mjs9-TY,6772
-torchzero/optim/__init__.py,sha256=vk6pIYJHWAGYJMdtJ1otsmVph-pdL5HwBg-CTeBCGso,253
-torchzero/optim/modular.py,sha256=B1ypLnbGY87nUdylPcbukdNoXvKa5GHCl-_14XRqLWs,6066
-torchzero/optim/experimental/__init__.py,sha256=RqNzJu5mVl3T0u7cf4TBzSiA20M1kxTZVYWjSVhEHuU,585
-torchzero/optim/experimental/experimental.py,sha256=tMHZVbEXm3s6mMr7unFSvk_Jks3uAaAG0fzsH6gr098,10928
-torchzero/optim/experimental/ray_search.py,sha256=GYyssL64D6RiImrZ2tchoZJ04x9rX-Bp1y2nQXEGxX0,2662
-torchzero/optim/first_order/__init__.py,sha256=CRT4farcwi8sO1qqDGxXv1856zOwuKlJKBIAIvpL2Z0,336
-torchzero/optim/first_order/cautious.py,sha256=XBeqrLQ4gFKVUYnJI5ROmF9wQJGhY90HR6UG5IS7vYk,6610
-torchzero/optim/first_order/forward_gradient.py,sha256=EM6W8MezS6iUtW36lxozdo2U4aqlDKE7Zf-0s1LACXQ,3066
-torchzero/optim/first_order/optimizers.py,sha256=jYmU6YDsYRGMRsCNkYc6AlvOf3wlU7Uv1xUrzl0o8zo,24501
-torchzero/optim/quasi_newton/__init__.py,sha256=0X83dl-85_j3ck8itWxJR49ZbFeOcWurW6FI8J12F1w,49
-torchzero/optim/quasi_newton/directional_newton.py,sha256=oZ-If8SRcFXTDFKS_zlAcJnif-v5dTCR9HXqmfvsvNA,2595
-torchzero/optim/second_order/__init__.py,sha256=3Gt0dR4NzBK07TV0NF8KZImUGHbI8E2zncDmhIC377I,31
-torchzero/optim/second_order/newton.py,sha256=-DqJrS8JPea6Y9jp5lDV14KyP8A24YKiPxJ32Vsfiv4,3848
-torchzero/optim/wrappers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-torchzero/optim/wrappers/nevergrad.py,sha256=4PLqfs2L9XJhveyX6l7kJu1cIPl6uv7_UD76amIlP7I,4733
-torchzero/optim/wrappers/nlopt.py,sha256=fGDOZ82sRI2VLH3hKIAhZY4EuKSdu_g217c-NZvD_rs,7104
-torchzero/optim/wrappers/scipy.py,sha256=_BQwFDQ7SBqIA5i1SJ29Xj0jDXVV8MQ_9RcsPT3U6VQ,18047
-torchzero/optim/zeroth_order/__init__.py,sha256=_6T0znO6V63Niq7DMhJPgUuMc_nPvAGxjCjMdf-r64U,218
-torchzero/optim/zeroth_order/fdm.py,sha256=5iJc_F_tRR4cGQfy2Jr8PmAnCGrPva89ZWczSdcBkFk,3686
-torchzero/optim/zeroth_order/newton_fdm.py,sha256=-5E1FGzeJMr8_IougzE_FEOPFt9pEjQxID4Y89Hpmh0,6537
-torchzero/optim/zeroth_order/rfdm.py,sha256=_Y7yiF1bsVRlXt5IK-3zQccwVl95JF0-Xw-fl8Q_7y4,10529
-torchzero/optim/zeroth_order/rs.py,sha256=3w2nnPGWPecourEdUG583vchcqdNxC6Q_PBL3l0PvCk,3333
-torchzero/random/__init__.py,sha256=8EowQhC4yTZuF8w1ZDl73YZtLej8SuhxCk1Bkifbkms,93
-torchzero/random/random.py,sha256=Oq4GvM_6AOsabg5ke6b8h51V9npyHVxp8ob_wC5D-Aw,2829
-torchzero/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-torchzero/utils/compile.py,sha256=pYEyX8P26iCb_hFqAXC8IP2SSQrRfC7ZDhXS0vVCsfY,1257
-torchzero/utils/derivatives.py,sha256=koLmuUcVcX41SrH_9rvfJyMXyHyocNLuZ-C8Kr2B7hk,4844
-torchzero/utils/python_tools.py,sha256=kkyDhoP695HhapfKrdjcrRbRAbcvB0ArP1pkxuVUlf0,1192
-torchzero/utils/torch_tools.py,sha256=sSBY5Bmk9LOAgPtaq-6TK4wDgPXsg6FIWxv8CVDx82k,3580
-torchzero-0.1.3.dist-info/LICENSE,sha256=r9ZciAoZoqKC_FNADE0ORukj1p1XhLXEbegdsAyqhJs,1087
-torchzero-0.1.3.dist-info/METADATA,sha256=AEBDfBalgFdxHAYaGaEV5SMtP_AOu915DzB5Kw0Qtho,6096
-torchzero-0.1.3.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-torchzero-0.1.3.dist-info/top_level.txt,sha256=isztuDR1ZGo8p2tORLa-vNuomcbLj7Xd208lhd-pVPs,10
-torchzero-0.1.3.dist-info/RECORD,,