torchzero 0.3.8__py3-none-any.whl → 0.3.10__py3-none-any.whl

torchzero/modules/experimental/algebraic_newton.py

@@ -1,145 +0,0 @@
-import warnings
-from functools import partial
-from typing import Literal
-from collections.abc import Callable
-import torch
-import torchalgebras as ta
-
-from ...core import Chainable, apply, Module
-from ...utils import vec_to_tensors, TensorList
-from ...utils.derivatives import (
-    hessian_list_to_mat,
-    hessian_mat,
-    jacobian_and_hessian_wrt,
-)
-
-class MaxItersReached(Exception): pass
-def tropical_lstsq(
-    H: torch.Tensor,
-    g: torch.Tensor,
-    solver,
-    maxiter,
-    tol,
-    algebra,
-    verbose,
-):
-    """it can run on any algebra with add despite it saying tropical"""
-    algebra = ta.get_algebra(algebra)
-
-    x = torch.zeros_like(g, requires_grad=True)
-    best_x = x.detach().clone()
-    best_loss = float('inf')
-    opt = solver([x])
-
-    niter = 0
-    def closure(backward=True):
-        nonlocal niter, best_x, best_loss
-        if niter == maxiter: raise MaxItersReached
-        niter += 1
-
-        g_hat = algebra.mm(H, x)
-        loss = torch.nn.functional.mse_loss(g_hat, g)
-        if loss < best_loss:
-            best_x = x.detach().clone()
-            best_loss = loss.detach()
-
-        if backward:
-            opt.zero_grad()
-            loss.backward()
-        return loss
-
-    loss = None
-    prev_loss = float('inf')
-    for i in range(maxiter):
-        try:
-            loss = opt.step(closure)
-            if loss == 0: break
-            if tol is not None and prev_loss - loss < tol: break
-            prev_loss = loss
-        except MaxItersReached:
-            break
-
-    if verbose: print(f'{best_loss = } after {niter} iters')
-    return best_x.detach()
-
-def tikhonov(H: torch.Tensor, reg: float, algebra: ta.Algebra = ta.TropicalSemiring()):
-    if reg!=0:
-        I = ta.AlgebraicTensor(torch.eye(H.size(-1), dtype=H.dtype, device=H.device), algebra)
-        I = I * reg
-        H = algebra.add(H, I.data)
-    return H
-
-
-class AlgebraicNewton(Module):
-    """newton in other algebras, not that it works."""
-    def __init__(
-        self,
-        reg: float | None = None,
-        hessian_method: Literal["autograd", "func", "autograd.functional"] = "autograd",
-        vectorize: bool = True,
-        solver=lambda p: torch.optim.LBFGS(p, line_search_fn='strong_wolfe'),
-        maxiter=1000,
-        tol: float | None = 1e-10,
-        algebra: ta.Algebra | str = 'tropical max',
-        verbose: bool = False,
-        inner: Chainable | None = None,
-    ):
-        defaults = dict(reg=reg, hessian_method=hessian_method, vectorize=vectorize)
-        super().__init__(defaults)
-
-        self.algebra = ta.get_algebra(algebra)
-        self.lstsq_args:dict = dict(solver=solver, maxiter=maxiter, tol=tol, algebra=algebra, verbose=verbose)
-
-        if inner is not None:
-            self.set_child('inner', inner)
-
-    @torch.no_grad
-    def step(self, vars):
-        params = TensorList(vars.params)
-        closure = vars.closure
-        if closure is None: raise RuntimeError('NewtonCG requires closure')
-
-        settings = self.settings[params[0]]
-        reg = settings['reg']
-        hessian_method = settings['hessian_method']
-        vectorize = settings['vectorize']
-
-        # ------------------------ calculate grad and hessian ------------------------ #
-        if hessian_method == 'autograd':
-            with torch.enable_grad():
-                loss = vars.loss = vars.loss_approx = closure(False)
-                g_list, H_list = jacobian_and_hessian_wrt([loss], params, batched=vectorize)
-                g_list = [t[0] for t in g_list] # remove leading dim from loss
-                vars.grad = g_list
-                H = hessian_list_to_mat(H_list)
-
-        elif hessian_method in ('func', 'autograd.functional'):
-            strat = 'forward-mode' if vectorize else 'reverse-mode'
-            with torch.enable_grad():
-                g_list = vars.get_grad(retain_graph=True)
-                H: torch.Tensor = hessian_mat(partial(closure, backward=False), params,
-                                method=hessian_method, vectorize=vectorize, outer_jacobian_strategy=strat) # pyright:ignore[reportAssignmentType]
-
-        else:
-            raise ValueError(hessian_method)
-
-        # -------------------------------- inner step -------------------------------- #
-        if 'inner' in self.children:
-            g_list = apply(self.children['inner'], list(g_list), params=params, grads=list(g_list), vars=vars)
-        g = torch.cat([t.view(-1) for t in g_list])
-
-        # ------------------------------- regulazition ------------------------------- #
-        if reg is not None: H = tikhonov(H, reg)
-
-        # ----------------------------------- solve ---------------------------------- #
-        tropical_update = tropical_lstsq(H, g, **self.lstsq_args)
-        # what now? w - u is not defined, it is defined for max version if u < w
-        # w = params.to_vec()
-        # w_hat = self.algebra.sub(w, tropical_update)
-        # update = w_hat - w
-        # no
-        # it makes sense to solve tropical system and sub normally
-        # the only thing is that tropical system can have no solutions
-
-        vars.update = vec_to_tensors(tropical_update, params)
-        return vars

torchzero/modules/experimental/tropical_newton.py

@@ -1,136 +0,0 @@
-import warnings
-from functools import partial
-from typing import Literal
-from collections.abc import Callable
-import torch
-
-from ...core import Chainable, apply, Module
-from ...utils import vec_to_tensors, TensorList
-from ...utils.derivatives import (
-    hessian_list_to_mat,
-    hessian_mat,
-    jacobian_and_hessian_wrt,
-)
-from ..second_order.newton import lu_solve, cholesky_solve, least_squares_solve
-
-def tropical_sum(x, dim): return torch.amax(x, dim=dim)
-def tropical_mul(x, y): return x+y
-
-def tropical_matmul(x: torch.Tensor, y: torch.Tensor):
-    # this imlements matmul by calling mul and sum
-
-    x_squeeze = False
-    y_squeeze = False
-
-    if x.ndim == 1:
-        x_squeeze = True
-        x = x.unsqueeze(0)
-
-    if y.ndim == 1:
-        y_squeeze = True
-        y = y.unsqueeze(1)
-
-    res = tropical_sum(tropical_mul(x.unsqueeze(-1), y.unsqueeze(-3)), dim = -2)
-
-    if x_squeeze: res = res.squeeze(-2)
-    if y_squeeze: res = res.squeeze(-1)
-
-    return res
-
-def tropical_dot(x:torch.Tensor, y:torch.Tensor):
-    assert x.ndim == 1 and y.ndim == 1
-    return tropical_matmul(x.unsqueeze(0), y.unsqueeze(1))
-
-def tropical_outer(x:torch.Tensor, y:torch.Tensor):
-    assert x.ndim == 1 and y.ndim == 1
-    return tropical_matmul(x.unsqueeze(1), y.unsqueeze(0))
-
-
-def tropical_solve(A: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
-    r = b.unsqueeze(1) - A
-    return r.amin(dim=-2)
-
-def tropical_solve_and_reconstruct(A: torch.Tensor, b: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
-    r = b.unsqueeze(1) - A
-    x = r.amin(dim=-2)
-    b_hat = tropical_matmul(A, x)
-    return x, b_hat
-
-def tikhonov(H: torch.Tensor, reg: float):
-    if reg!=0: H += torch.eye(H.size(-1), dtype=H.dtype, device=H.device) * reg
-    return H
-
-
-class TropicalNewton(Module):
-    """suston"""
-    def __init__(
-        self,
-        reg: float | None = None,
-        hessian_method: Literal["autograd", "func", "autograd.functional"] = "autograd",
-        vectorize: bool = True,
-        interpolate:bool=False,
-        inner: Chainable | None = None,
-    ):
-        defaults = dict(reg=reg, hessian_method=hessian_method, vectorize=vectorize, interpolate=interpolate)
-        super().__init__(defaults)
-
-        if inner is not None:
-            self.set_child('inner', inner)
-
-    @torch.no_grad
-    def step(self, vars):
-        params = TensorList(vars.params)
-        closure = vars.closure
-        if closure is None: raise RuntimeError('NewtonCG requires closure')
-
-        settings = self.settings[params[0]]
-        reg = settings['reg']
-        hessian_method = settings['hessian_method']
-        vectorize = settings['vectorize']
-        interpolate = settings['interpolate']
-
-        # ------------------------ calculate grad and hessian ------------------------ #
-        if hessian_method == 'autograd':
-            with torch.enable_grad():
-                loss = vars.loss = vars.loss_approx = closure(False)
-                g_list, H_list = jacobian_and_hessian_wrt([loss], params, batched=vectorize)
-                g_list = [t[0] for t in g_list] # remove leading dim from loss
-                vars.grad = g_list
-                H = hessian_list_to_mat(H_list)
-
-        elif hessian_method in ('func', 'autograd.functional'):
-            strat = 'forward-mode' if vectorize else 'reverse-mode'
-            with torch.enable_grad():
-                g_list = vars.get_grad(retain_graph=True)
-                H: torch.Tensor = hessian_mat(partial(closure, backward=False), params,
-                                method=hessian_method, vectorize=vectorize, outer_jacobian_strategy=strat) # pyright:ignore[reportAssignmentType]
-
-        else:
-            raise ValueError(hessian_method)
-
-        # -------------------------------- inner step -------------------------------- #
-        if 'inner' in self.children:
-            g_list = apply(self.children['inner'], list(g_list), params=params, grads=list(g_list), vars=vars)
-        g = torch.cat([t.view(-1) for t in g_list])
-
-        # ------------------------------- regulazition ------------------------------- #
-        if reg is not None: H = tikhonov(H, reg)
-
-        # ----------------------------------- solve ---------------------------------- #
-        tropical_update, g_hat = tropical_solve_and_reconstruct(H, g)
-
-        g_norm = torch.linalg.vector_norm(g) # pylint:disable=not-callable
-        abs_error = torch.linalg.vector_norm(g-g_hat) # pylint:disable=not-callable
-        rel_error = abs_error/g_norm.clip(min=1e-8)
-
-        if interpolate:
-            if rel_error > 1e-8:
-
-                update = cholesky_solve(H, g)
-                if update is None: update = lu_solve(H, g)
-                if update is None: update = least_squares_solve(H, g)
-
-                tropical_update.lerp_(update.ravel(), rel_error.clip(max=1))
-
-        vars.update = vec_to_tensors(tropical_update, params)
-        return vars

torchzero-0.3.8.dist-info/RECORD

@@ -1,130 +0,0 @@
-docs/source/conf.py,sha256=jd80ZT2IdCx7nlQrpOTJL8UhGBNm6KYyXlpp0jmRiAw,1849
-tests/test_identical.py,sha256=NZ7A8Rm1U9Q16d-cG2G_wccpPtNALyoKYJt9qMownMc,11568
-tests/test_module.py,sha256=qX3rjdSJsbA8JO17bPTUIDspe7bg2dogqxMw__KV7SU,2039
-tests/test_opts.py,sha256=XfpDaVwOC2VuG700BXWAFWiemeVW0ucLG74yfns9mB8,40849
-tests/test_tensorlist.py,sha256=VWX9wYdfkG-0Y8I0wWPp56ZJM0mBNPvS_SC3irmcYcs,72427
-tests/test_utils_optimizer.py,sha256=bvC0Ehvs2L8fohpyIF5Vfr9OKTycpnODWLPflXilU1c,8414
-tests/test_vars.py,sha256=3p9dsHk7SJpMd-WRD0ziBNq5FEHRBJGSxbMLD8ES4J0,6815
-torchzero/__init__.py,sha256=L7IJ1qZ3o8E9oRwlJZBK2_2yII_eeGEk57Of6EfVbrk,112
-torchzero/core/__init__.py,sha256=2JRyeGZprTexAeEPQOIl9fLFGBwzvya-AwKyt7XAmGQ,210
-torchzero/core/module.py,sha256=Razw3c71Kfegznm0vQxsii1KuTUCPBC9UGyq2v-KX4M,27568
-torchzero/core/preconditioner.py,sha256=R1IGk7Tbea5wSkazpnXwusjvBxzJHzEWgCtR_nEz2w4,6258
-torchzero/core/transform.py,sha256=ajNJcX45ds-_lc5CqxgLfEFGil6_BYLerB0WvoTi8rM,10303
-torchzero/modules/__init__.py,sha256=BDeyuSd2s1WFUUXIo3tGTNp4aYp4A2B94cydpPW24nY,332
-torchzero/modules/functional.py,sha256=HXNzmPe7LsPadryEm7zrcEKqGej16QDwSgBkbEvggFM,6492
-torchzero/modules/clipping/__init__.py,sha256=ZaffMF7mIRK6hZSfuZadgjNTX6hF5ANiLBny2w3S7I8,250
-torchzero/modules/clipping/clipping.py,sha256=I-5utyrqdKtF5yaH-9m2F3UqdfpPmA2bSSFUAZ_d60Q,12544
-torchzero/modules/clipping/ema_clipping.py,sha256=pLeNuEBLpJ74io2sHn_ZVYaQ6ydEfhpVfVEX2bFttd0,5947
-torchzero/modules/clipping/growth_clipping.py,sha256=OD-kdia2Rn-DvYlYV6EZlGPDVTh9tj-W9mpiZPc3hOQ,6772
-torchzero/modules/experimental/__init__.py,sha256=fEPDYDl7qhaFoferDRmG3ehwuqSvx4Vt2uOz0Y7h4to,483
-torchzero/modules/experimental/absoap.py,sha256=Z4MS4pDPSQ9IaTk8g57OfrsWcYVOT72x533KKtn2Zxk,13512
-torchzero/modules/experimental/adadam.py,sha256=OAPF1-NUbg79V3QOTYzsQlRC97C7XHj5boOLDqLz3PE,4029
-torchzero/modules/experimental/adamY.py,sha256=g1pAHwgdyDdKvObZ67lCSc36L99tl5jlQgOr4lMJCDo,4595
-torchzero/modules/experimental/adasoap.py,sha256=PduidICEGYICIvlYysYCFZF7-QhNX0YlhHfPhLONnUs,11247
-torchzero/modules/experimental/algebraic_newton.py,sha256=sq5ZD_j_EtlxIjNnS0rKKwTSG_JuwsZOg9ZMMQTuQm0,5154
-torchzero/modules/experimental/curveball.py,sha256=Uk30uLEztTHD5IUJLJm9Nn3x31DF9kQHmeLFhc065us,3262
-torchzero/modules/experimental/gradmin.py,sha256=iJmEvDEdVdck0C-94pY3iGxnIoNv6Fu6vj3f7lS6aQM,3686
-torchzero/modules/experimental/newton_solver.py,sha256=iGI2LHLaZd2ovpbq1Vogs76os0zWG7VwM7nUz8RzxVg,3071
-torchzero/modules/experimental/reduce_outward_lr.py,sha256=kjtRwepBGBca77ToM-lw3b8ywptMtmSdC_jQfjJAwlY,1184
-torchzero/modules/experimental/soapy.py,sha256=Ishd2Jj6BbhjrLyC48zf-cjMmA1kJb_uKXESQBIML_s,10990
-torchzero/modules/experimental/spectral.py,sha256=8_n208V2yPY3z5pCym-FvwO7DGFhozNgWlpIBtQSdrI,12139
-torchzero/modules/experimental/subspace_preconditioners.py,sha256=WnHpga7Kx4-N2xU5vP3uUHRER70ymyNJCWbSx2zXWOk,4976
-torchzero/modules/experimental/tropical_newton.py,sha256=uq66ouhgrgc8iYGozDQ3_rtbubj8rKRwb1jfcdnlpHg,4903
-torchzero/modules/grad_approximation/__init__.py,sha256=DVFjf0cXuF70NA0nJ2WklpP01PQgrRZxUjUQjjQeSos,195
-torchzero/modules/grad_approximation/fdm.py,sha256=2PNNBIMup1xlOwLFAwAS3xAVd-7GGVyerMeKH1ug9LQ,3591
-torchzero/modules/grad_approximation/forward_gradient.py,sha256=Kb8RNGAIb2tKzgofnEn4pQjS7TPq824B_P14idyy8e0,3564
-torchzero/modules/grad_approximation/grad_approximator.py,sha256=Pa1Lv52T7WawUJUUA3IHm7mVypBQXLbjc5_15FkVwnQ,2938
-torchzero/modules/grad_approximation/rfdm.py,sha256=s7OSMFnIEr43WKCT0TXdgzz_6odOkRN0BcKWkFbbPAE,10189
-torchzero/modules/line_search/__init__.py,sha256=nkOUPLe88wE91ICEhprl2pJsvaKtbI3KzYOdT83AGsg,253
-torchzero/modules/line_search/backtracking.py,sha256=FG_-KAN9whvBNZyhDa5-ta46IQFm8hagVvaPTXCCV88,6307
-torchzero/modules/line_search/line_search.py,sha256=4z0fHJAGAZT2IVAOUxZetAszPtNuXfXdFzs1_WUWT2c,7296
-torchzero/modules/line_search/scipy.py,sha256=7tfxXT8RAIHpRv-e5w9C8RNvkvgwgxHZaWI25RjTYy0,1156
-torchzero/modules/line_search/strong_wolfe.py,sha256=Y6UXd2Br30YWta1phZx1wiSsFQC6wbgmvOpVITcmJpw,7504
-torchzero/modules/line_search/trust_region.py,sha256=_zOje00BLvIMi0d5H9qZavqf3MWeB48Q-WosgXu3Ef4,2349
-torchzero/modules/lr/__init__.py,sha256=pNxbBUGzDp24O6g7pu1bRg1tzh4eh-mSxVbhOItKHpc,90
-torchzero/modules/lr/lr.py,sha256=wlubixzgxnm4ucyiEtGWzQOskaLXLInvSaR0sGKxto8,2161
-torchzero/modules/lr/step_size.py,sha256=0HWYAYhVqWCCYe_-guBnMaOpqLbsMm4-F6bRFjltBsc,4036
-torchzero/modules/momentum/__init__.py,sha256=pSD7vxu8PySrYOSHQMi3C9heYdcQr8y6WC_rwMybZm0,544
-torchzero/modules/momentum/averaging.py,sha256=hyH5jzvYTbB1Vcjx0j_v4dtPp54GUUDOZYVDADGjcfE,2672
-torchzero/modules/momentum/cautious.py,sha256=QCoBXpYcIUOrgY6XXHA30m0-MVy7iGCGxZGFLyDwqkc,5841
-torchzero/modules/momentum/ema.py,sha256=4ubPpq9TL0oQZ5_eXBwU5oRbxV3faHMEM1a_kv8vRqI,7733
-torchzero/modules/momentum/experimental.py,sha256=ze9oxqxdmqRFQyVdG7iBA-hICft5mxeAM6GCTQ4ewes,6352
-torchzero/modules/momentum/matrix_momentum.py,sha256=IQjCp2Kb53bCaReM7fHBil_pwH9oiH029YkWFq0OIDw,4894
-torchzero/modules/momentum/momentum.py,sha256=hcmmYysGItb3b7MBBVhoODh7p4Fyit68cZzD0NUBmvA,1540
-torchzero/modules/ops/__init__.py,sha256=hxMZFSXX7xvitXkuBiYykVGX3p03Xprm_QA2CMg4eW8,1601
-torchzero/modules/ops/accumulate.py,sha256=YGI11YxgTWvIBq5maDRWiSA-v-FS-XoaSYPU2SSrBY8,2759
-torchzero/modules/ops/binary.py,sha256=-b0yvKvfDx9-HcaaxLWzg5C6rUl24oP3OltSF-iXi6w,9731
-torchzero/modules/ops/debug.py,sha256=9sJOHRMwTMaOgOi2QFwCH7g2WPF1o3oyouPJO-MQQg4,862
-torchzero/modules/ops/misc.py,sha256=xdxnGbRArWBqzyufUdrCQH-mAI9utRF0zxcvWCkEfZc,16383
-torchzero/modules/ops/multi.py,sha256=P7mSG0LnDMkuZNSgtpHRNgqglqksrdxITCzkhmEjqxU,5742
-torchzero/modules/ops/reduce.py,sha256=xvFHZG5Wf7KxfFLkynFGBOK6xywyTXsbCasW6h2OYAU,5695
-torchzero/modules/ops/split.py,sha256=fFcDnJZ-e46cx_fx_TkGlVsFYOL1Y8UAp_pUPJOOdm4,2303
-torchzero/modules/ops/switch.py,sha256=5idKd9xBP-KbqZjWBcr6ZDjso8BRpTNQYJg4xKWwmng,2511
-torchzero/modules/ops/unary.py,sha256=h3MXS6jydZjfFetjaBCWCUWTXdQcNKnxEC6uGS6yh3c,4794
-torchzero/modules/ops/utility.py,sha256=p-mc2j1mQEMLxp4brnAnzgmK6VKbSnYd2U8vkAwTKd8,3117
-torchzero/modules/optimizers/__init__.py,sha256=BbT2nhIt4p74t1cO8ziQgzqZHaLvyuleXQbccugd06M,554
-torchzero/modules/optimizers/adagrad.py,sha256=1DIBJ_7gJ35qidXMK4IkHYF_37Bl9Ptl9mAgfOq6YAk,4834
-torchzero/modules/optimizers/adam.py,sha256=xctnENJ9rcpv2sis4zAGPGoy-ccJC1iVl8SvBynaG50,4093
-torchzero/modules/optimizers/lion.py,sha256=eceNfITCozqYob0thWbIV7AdY1yAIJMqb4GJfB8a1SA,1087
-torchzero/modules/optimizers/muon.py,sha256=m3LpwD6AF7E-1v3VVPHAN8S_tPTTFKZ5RpkzKea4K4g,9598
-torchzero/modules/optimizers/orthograd.py,sha256=5BLnNJTYuGUClHmlxaXZ1jNvBR4zSFDGG8nM20lZdhk,2046
-torchzero/modules/optimizers/rmsprop.py,sha256=d10Y9Ck-391tVysO3xMHg3g2Pe0UEZplgebEyDYi3Z4,4333
-torchzero/modules/optimizers/rprop.py,sha256=n4k5-9F3ppH0Xl-4l4vNXfqVf2r67vMPCkstUaQKPLw,10974
-torchzero/modules/optimizers/shampoo.py,sha256=AHHV6d71DqKDPCg52ShWIPIRSGtWkMc1v1XwXgDG3qY,8606
-torchzero/modules/optimizers/soap.py,sha256=Kf2BAtIf2QY1V2ZJcUjRLcp2WfIVLd3mNclnaT3Nmds,11520
-torchzero/modules/optimizers/sophia_h.py,sha256=8pSlYVm66xWplzdP8MX3MCTzzIYHsxGzDEXJKA03Zgg,4279
-torchzero/modules/projections/__init__.py,sha256=OCxlh_-Tx-xpl31X03CeFJH9XveH563oEsWc8rUvX0A,196
-torchzero/modules/projections/dct.py,sha256=wxaEV6dTNiOqW_n2UHX0De6mMXTKDXK6UNcMNI4Rogk,2373
-torchzero/modules/projections/fft.py,sha256=OpCcEM1-A2dgk1umwRsBsvK7ObiHtsBKlkkcw0IX83Q,2961
-torchzero/modules/projections/galore.py,sha256=c9CZ0kHxpKEoyfc_lnmeHOkNp55jCppb7onN5YmWnN8,242
-torchzero/modules/projections/projection.py,sha256=aYufSD3ftRUqVScPmqxwEFgP1P8ioxM8z9eyzaL7d10,10147
-torchzero/modules/projections/structural.py,sha256=QaCGHmzHCXj46sM-XZ5XlYU9BnuRKI2ReR3LE8y2R4g,5740
-torchzero/modules/quasi_newton/__init__.py,sha256=0iOlX73PHj9lQS3_2cJ5lyCdas904MnFfIvR8Popvzw,402
-torchzero/modules/quasi_newton/cg.py,sha256=h-di1oKKP1tDoh-LogBRIRCp2UF9GA6XjEJPlX6xXf4,9322
-torchzero/modules/quasi_newton/lbfgs.py,sha256=U7FKNqFTRdabf1_UUCCEO3JoDlOnWzGBhYNvVg138gg,9199
-torchzero/modules/quasi_newton/lsr1.py,sha256=BuoztcRo0lm7WW3vKtDQcfKizF-9WPItOY_X9Ng1ZS8,6033
-torchzero/modules/quasi_newton/olbfgs.py,sha256=2YAOXlMnPGw22sNcIMH1hmggzAXQRbN59RSPUZNKUZY,8352
-torchzero/modules/quasi_newton/quasi_newton.py,sha256=5FutzdBNpx6P8Qun9LjXd-rsy2nY2bkpQ0z0cLPnbJo,17373
-torchzero/modules/quasi_newton/experimental/__init__.py,sha256=3qpZGgdsx6wpoafWaNWx-eamRl1FuxVCWQZq8Y7Cl98,39
-torchzero/modules/quasi_newton/experimental/modular_lbfgs.py,sha256=ec6JKYX89xA_UlY9VrMB3hBjDyNKwkalS_4JQGA1qOY,10762
-torchzero/modules/second_order/__init__.py,sha256=jolCGaIVkID9hpxgx0Tc22wgjVlwuWekWjKTMe5jKXw,114
-torchzero/modules/second_order/newton.py,sha256=XNhscAuWwxOUwps3sUrxc2ExgkNFbilnAdszrCvQxFg,5845
-torchzero/modules/second_order/newton_cg.py,sha256=stVySgo7tmvntd-tuAzThzpWmZzfTnmn8ISQa5Oi4yw,2872
-torchzero/modules/second_order/nystrom.py,sha256=ZyCWrde-_-Ednj46jafuvBOzG3nC-3cPYGr-HytZbsE,6073
-torchzero/modules/smoothing/__init__.py,sha256=tUTGN0A-EQC7xuLV2AuHFWk-t7D6jIJlpV_3qyfRqLk,80
-torchzero/modules/smoothing/gaussian.py,sha256=YlT_G4MqAVkiWG56RHAwgt5SSPISpvQZQbSLh8mhF3I,6153
-torchzero/modules/smoothing/laplacian.py,sha256=Bfrs7D59SfdU7j-97UBKD1hs0obC-ZgjJvG7oKwaa0o,5065
-torchzero/modules/weight_decay/__init__.py,sha256=VdJfEx3uk8wYGCpMjYSeudXyGX8ONqsQYoBCE3cdM1U,72
-torchzero/modules/weight_decay/weight_decay.py,sha256=p6jGD3hgC_rmZXiWYr7_IZWHMdVJJaT_bcHHzcdXSxU,1912
-torchzero/modules/wrappers/__init__.py,sha256=6b5Ac-8u18IVp_Jnw1T1xQExwpQhpQ0JwNV9GyC_Yj8,31
-torchzero/modules/wrappers/optim_wrapper.py,sha256=mcoQCUJwpMJuCDv03nDa0jZIb3Y0CyaeE1kNcJQozfo,3582
-torchzero/optim/__init__.py,sha256=aXf7EkywqYiR50I4QeeVXro9aBhKiqfbY_BCia59sgU,46
-torchzero/optim/utility/__init__.py,sha256=pUacok4XmebfxofE-QWZLgViajsU-3JkXcWi9OS-Jrw,24
-torchzero/optim/utility/split.py,sha256=ZbazNuMTYunm75V_5ard0A_LletGaYAg-Pm2rANJKrE,1610
-torchzero/optim/wrappers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-torchzero/optim/wrappers/nevergrad.py,sha256=2jHWQiWGjaffAqhJotMwOt03OtW-L57p8OesD2gVVow,3949
-torchzero/optim/wrappers/nlopt.py,sha256=ZoHBf51OhwgAaExxmoFtvP8GqO9uBHdEsc4HLm0wcic,7588
-torchzero/optim/wrappers/scipy.py,sha256=0BNBlHCbeTslXkXhnKvhuvJfNO7_CHFa2AXruYySnzM,14561
-torchzero/utils/__init__.py,sha256=By___ngB1bcnrSZiJanvtKk8QFrPmLRhTOrkFYP2MU4,929
-torchzero/utils/compile.py,sha256=N8AWLv_7oBUHYornmvvx_L4uynjiD-x5Hj1tBwei3-w,5127
-torchzero/utils/derivatives.py,sha256=S4Vh2cwE2h6yvhqu799AjR4GVHOEg7yApH3SataKxnA,16881
-torchzero/utils/numberlist.py,sha256=cbG0UsSb9WCRxVhw8sd7Yf0bDy_gSqtghiJtkUxIO6U,6139
-torchzero/utils/ops.py,sha256=n4Su1sbgTzlHczuPEHkuWenTtNBCa_MvlQ_hCZkIPnQ,314
-torchzero/utils/optimizer.py,sha256=-vuOZNu4luSZA5YtwC_7s-G2FvHKnM2k5KqC6bC_hcM,13097
-torchzero/utils/optuna_tools.py,sha256=F-1Xg0n_29MVEb6lqgUFFNIl9BNJ6MOdIJPduoNH4JU,1325
-torchzero/utils/params.py,sha256=nQo270aOURU7rJ_D102y2pSXbzhJPK0Z_ehx4mZBMes,5784
-torchzero/utils/python_tools.py,sha256=RFBqNj8w52dpJ983pUPPDbg2x1MX_-SsBnBMffWGGIk,2066
-torchzero/utils/tensorlist.py,sha256=qSbiliVo1euFAksdHHHRbPUdYYxfkw1dvhpXj71wGy0,53162
-torchzero/utils/torch_tools.py,sha256=ohqnnZRlqdfp5PAfMSbQDIEKygW0_ARjxSEBp3Zo9nU,4756
-torchzero/utils/linalg/__init__.py,sha256=Dzbho3_z7JDdKzYD-QdLArg0ZEoC2BVGdlE3JoAnXHQ,272
-torchzero/utils/linalg/benchmark.py,sha256=wiIMn-GY2xxWbHVf8CPbJddUPeUPq9OUDkvbp1iILYI,479
-torchzero/utils/linalg/matrix_funcs.py,sha256=-LecWrPWbJvfeCgIzUhfWARa2aSZvJ12lHX7Jno38O4,3099
-torchzero/utils/linalg/orthogonalize.py,sha256=mDCkET7qgDZqf_y6oPYAK3d2L5HrB8gzOFPl0YoONaY,399
-torchzero/utils/linalg/qr.py,sha256=L-RXuYV-SIHI-Llq4y1rQ_Tz-yamds0_QNZeHapbjNE,2507
-torchzero/utils/linalg/solve.py,sha256=P0PMi0zro3G3Rd0X-JeoLk7tqYDB0js0aB4bpQ0OABU,5235
-torchzero/utils/linalg/svd.py,sha256=wBxl-JSciINV-N6zvM4SGdveqMr6idq51h68LyQQRYg,660
-torchzero-0.3.8.dist-info/licenses/LICENSE,sha256=r9ZciAoZoqKC_FNADE0ORukj1p1XhLXEbegdsAyqhJs,1087
-torchzero-0.3.8.dist-info/METADATA,sha256=vj5aue0pVG8xNStpOEvPfln422K5fpV3BKF-H7ZlhRQ,13941
-torchzero-0.3.8.dist-info/WHEEL,sha256=zaaOINJESkSfm_4HQVc5ssNzHCPXhJm0kEUakpsEHaU,91
-torchzero-0.3.8.dist-info/top_level.txt,sha256=YDdpIOb7HyKV9THOtOYsFFMTbxvCO0kiol4-83tDj-A,21
-torchzero-0.3.8.dist-info/RECORD,,