torchzero 0.3.14__py3-none-any.whl → 0.4.0__py3-none-any.whl

torchzero-0.4.0.dist-info/RECORD

@@ -0,0 +1,191 @@
+tests/test_identical.py,sha256=1_Rv-nwoVL1YcXPDA_DprjbT4jkvr0apUPbeQpebMUI,11508
+tests/test_module.py,sha256=qX3rjdSJsbA8JO17bPTUIDspe7bg2dogqxMw__KV7SU,2039
+tests/test_module_autograd.py,sha256=cncOlxtxmyJQHUd7nL9aWLRAr1kxtlKgVLqP3_qIb2E,21374
+tests/test_objective.py,sha256=HY0rK0z6PpiXvEsCu4mLgTlSVKusnT69S2GbuVcwMRo,7119
+tests/test_opts.py,sha256=2KqkvZhisKGmW2wm0d0K1gHjD3IU4V3Z0-6iNVEZ43M,43900
+tests/test_tensorlist.py,sha256=B0Tq4_r-1DOYpS360X7IsLQiWn5fukhIMDKZM6zVO2Y,72164
+tests/test_utils_optimizer.py,sha256=_JoMqvXXZ6TxugS_CmfmP55Vvp0XrSPCjSz2nJJmaoI,8399
+torchzero/__init__.py,sha256=iZFYBgpDDk4d-vunJR1TFqBAHKDAit_SHtepO0RRrIo,131
+torchzero/core/__init__.py,sha256=b6h0FXF9CQgXgaQY6rz_sjjs8zswLh4PZu5FQQqlLUo,339
+torchzero/core/chain.py,sha256=dtFpxnw8vcbi3EeAANXyPtUmyPyv_VuZrTiPlLRmh7c,1899
+torchzero/core/functional.py,sha256=X3igEP-F2_A6gDza8Gc8l3RcjMuuI4TEwmWA6e6NhKo,3154
+torchzero/core/modular.py,sha256=VGe9PHxOBxnJmMx3_JWFVqq5__xmxqXgQMrfbVJGVa4,9574
+torchzero/core/module.py,sha256=E08ZJnZGNlnojURk8Ztm7w-7Q1Qzq9aEI6NwuvC7OBA,18008
+torchzero/core/objective.py,sha256=6GYQrUDmjufvBJ9JByqnMXFpnjrZY2-mVDnlXRot_HU,39978
+torchzero/core/reformulation.py,sha256=UyAS_xq5sy_mMpmkvtwpHrZHd6Y2RgyPwN0zZlyxFTI,3857
+torchzero/core/transform.py,sha256=4JBpzDPBoKk3QWKm2Z4W6r49rh0UxdsoQjGMLbCmcFs,12245
+torchzero/linalg/__init__.py,sha256=K-9kRBysUd9HEeGfFAu5IbvTvm4xmRpQSu5k_EotYQE,347
+torchzero/linalg/benchmark.py,sha256=wiIMn-GY2xxWbHVf8CPbJddUPeUPq9OUDkvbp1iILYI,479
+torchzero/linalg/eigh.py,sha256=Dz3hz10u84mOEYKBp902byzYvn2b6kiDFBCxYzXhIWo,1409
+torchzero/linalg/linalg_utils.py,sha256=1RBpQevb25_hbRieONw6CgnoWOJFXXv8oWTMugEivEk,385
+torchzero/linalg/linear_operator.py,sha256=mVEOvu6yY7TYhUdmZm1IAc6_pWnTaykKDgZu_-J-atk,16653
+torchzero/linalg/matrix_power.py,sha256=gEWGvh3atc7745dwNcxNg0RtUrVgeKD6KxyRckKkkdQ,1255
+torchzero/linalg/orthogonalize.py,sha256=RVNoBRu4pXCPAYDV9EFIqsrwePNObKvavIdVTl-1eTU,3513
+torchzero/linalg/qr.py,sha256=v5jNT8BTwQGitdLA8xqOSxmK3fUBHksuvYBeRcscwfo,2541
+torchzero/linalg/solve.py,sha256=Jd7pgEEjxe1h5YQxKOhR8uwCjqCK2f1PY2KEHHUzW84,14496
+torchzero/linalg/svd.py,sha256=lvR8FpLk0H795i6DS4BtBZ1hYpm364JJOrGhOX62pEQ,694
+torchzero/linalg/torch_linalg.py,sha256=brhMXUZyYuxuEV-FyQerep1iL7auutW5kmgJpOzUROw,6001
+torchzero/modules/__init__.py,sha256=k79l7dMEmfxvikxfG8iUinsvtkngErSgKhHV3uqzLyc,587
+torchzero/modules/functional.py,sha256=aj7xqHmeze4izxG9k3L6ziG-K_yj8n8fkFpIv-X8V78,8141
+torchzero/modules/adaptive/__init__.py,sha256=pZVcpL3K3GqpfOHAqcQ6YV0EMib2587NY7FhESU671E,943
+torchzero/modules/adaptive/adagrad.py,sha256=AO7-DqibgfTDXDX11F0sjvfrm668ReMOQCG7HBT8TNY,12031
+torchzero/modules/adaptive/adahessian.py,sha256=2-7n412iK3aN0TvkxFao0nRNLJqlzxrlsfHsNiDkTg8,8730
+torchzero/modules/adaptive/adam.py,sha256=26CB7ja80XKjySIJ8BydnwVAr8qZyV68YFN-CI4el2g,3768
+torchzero/modules/adaptive/adan.py,sha256=XQuVkD1y8a0jmu-A7eHw5d8a6g6Bt0dFOLhRD6pggIg,4108
+torchzero/modules/adaptive/adaptive_heavyball.py,sha256=IguOBiYkBuCPacCFfgFP6oWF9lOS1_931h97PG1N13s,2247
+torchzero/modules/adaptive/aegd.py,sha256=WLN6vvbSRhQ1P753M3bx_becSF-3cTbu37nhz3NvdGM,1903
+torchzero/modules/adaptive/esgd.py,sha256=gQXUCBopPp5KVi2_nwt5NqoyQe6JOyoxcGbKp_OYma0,6169
+torchzero/modules/adaptive/lion.py,sha256=g8JFR_iueSxdSqAmDH4QLgeJ7cRTfarb-Te6zVobzQs,1050
+torchzero/modules/adaptive/lmadagrad.py,sha256=plE1FaB7JN9QqizwGtjzNRNDpxp_qqw1Z6ULUl4yzmA,9086
+torchzero/modules/adaptive/mars.py,sha256=_K2Agi5nC-H-jY0vbBwIMTtvbKUXu341X4YhDpCGQHg,2257
+torchzero/modules/adaptive/matrix_momentum.py,sha256=KBz6fCFVvLH_a99wfxUIvvtBWqm7xq3rToW2sz2vKbU,6790
+torchzero/modules/adaptive/msam.py,sha256=GAOeAQojx4k-cGrB-nN15YQBp_AJknPj4jStgGabcDk,6941
+torchzero/modules/adaptive/muon.py,sha256=CHclVyM7XLEVHjBGPI3a8Wg2S_xcOclr2IOQ4HstZOE,7728
+torchzero/modules/adaptive/natural_gradient.py,sha256=N9pYJcm0i7o-GlQIxs-oHRyhzJxQqtAiI_cIwKovpOc,6582
+torchzero/modules/adaptive/orthograd.py,sha256=0u2sfGZJjlJItLX2WRP5fLAD8Wd9SgJzQYAUpARJ64A,1813
+torchzero/modules/adaptive/rmsprop.py,sha256=qWVkRmUQ3dui9yBVYtAEll7OlXZDKNT_m70FakTOrTY,4529
+torchzero/modules/adaptive/rprop.py,sha256=HEZaZXo50HX1W7-X_e5Qhi2wk_d9wOFi1_BrKBvE6mo,11536
+torchzero/modules/adaptive/sam.py,sha256=PdcmxxgtTqGtbkHEA6d6JFy_cIqFVVA2XNxIcXQxvzc,5701
+torchzero/modules/adaptive/shampoo.py,sha256=ld_Nwoh2Hemt5QVB7BdKtaOczdIF1gK_MCiEhqwUnFI,8891
+torchzero/modules/adaptive/soap.py,sha256=cPRkgknFf26HfFss5YQ_RkLBdfy-nOvzFi7agVji3nk,12406
+torchzero/modules/adaptive/sophia_h.py,sha256=3l91pYvJBK1pdNodgYsTTR1cMnkqe6zePWSYSx-CAXQ,7059
+torchzero/modules/clipping/__init__.py,sha256=ZaffMF7mIRK6hZSfuZadgjNTX6hF5ANiLBny2w3S7I8,250
+torchzero/modules/clipping/clipping.py,sha256=_VY78EmzZiYCirZZ99cCSnim5wNgQyeXaXAGxPQUnFk,14184
+torchzero/modules/clipping/ema_clipping.py,sha256=D4NgXzXYMjK_SKQU3rVoOKzaCd9igGQg_7sXiGMgMqI,6750
+torchzero/modules/clipping/growth_clipping.py,sha256=I1nk5xXBjk0BzWYzMC58LZHouY44myZNIUjM-duv7zc,6508
+torchzero/modules/conjugate_gradient/__init__.py,sha256=G5WcVoiQYupRBeqjI4lCraGeXNSvWT-_-ynpcE6NQS8,184
+torchzero/modules/conjugate_gradient/cg.py,sha256=YQD01wNnSpUdtqcf7egXHpbhFhRtuFD8ZhlNLvc_ZI0,14539
+torchzero/modules/experimental/__init__.py,sha256=nI6WD-EDFfIav8G2mZuAznErweAqZ5C15qFKqY49ci8,728
+torchzero/modules/experimental/coordinate_momentum.py,sha256=HzKy8X5qEvud-xKHJYHpzH6ObxzvYcMcdgodsCw4Bbk,1099
+torchzero/modules/experimental/curveball.py,sha256=beHGD1Wh9GxYqMBh1k9Ru6TG3U9eZR6_l8ZUQcZzYxw,2765
+torchzero/modules/experimental/dct.py,sha256=CW-Y2gcjlHlxtIx7SekUOfw2EzujA6v0LcjDYGAfh6M,2433
+torchzero/modules/experimental/fft.py,sha256=s95EzvK4-ZJdwZbVhtqwirY9eVy7v6mFDRMgoLY9wjo,3020
+torchzero/modules/experimental/gradmin.py,sha256=LajM0GU1fB6PsGDg8k0KjKI73RvyZYqPvzcdoVYDq-c,3752
+torchzero/modules/experimental/higher_order_newton.py,sha256=qLSCbkmd7dw0lAhOJGpvvOesZfCMNt2Vz_mc7HknCMQ,12131
+torchzero/modules/experimental/l_infinity.py,sha256=nhYusM3YYbc0ptaSf9zlrsqY8EgxlHm9OejJ6VV0qtM,4750
+torchzero/modules/experimental/newton_solver.py,sha256=nsknGw8XDHKyDNE2cFYxRI5qF_mw0vragPtmzTY4-aM,4069
+torchzero/modules/experimental/newtonnewton.py,sha256=ryJV5eDqj4mEvIxERJIPd-0f57M0eFevH4ZAVHhnP4A,3892
+torchzero/modules/experimental/reduce_outward_lr.py,sha256=ehctg5zLEOHPfiQQUq5ShMj3pDhtxqdNUEneMR9l7Bs,1275
+torchzero/modules/experimental/scipy_newton_cg.py,sha256=psllNtDwUbkVAXBDKwWEueatOmDNPFy-pMwBkqF3_r0,3902
+torchzero/modules/experimental/spsa1.py,sha256=DiQ_nHAC8gnqoNNK7oe6djOiwpwvI5aPtpKA43F7jrQ,3607
+torchzero/modules/experimental/structural_projections.py,sha256=IwpgibNDO0slzMyi6djQXRhQO6IagUgUUCr_-7US1IE,4104
+torchzero/modules/grad_approximation/__init__.py,sha256=_mQ2sWvnMfqc3RQcVmZuBlphtLZCO7z819abGY6kYuM,196
+torchzero/modules/grad_approximation/fdm.py,sha256=zx70GZDQmhe43bZP5Mbbl31xsMOsGO43kznoQDbqxJo,4372
+torchzero/modules/grad_approximation/forward_gradient.py,sha256=7fKZoKetYzgD85L3W0x1oG56SdWHj5MDWwmWpV7bpr4,3949
+torchzero/modules/grad_approximation/grad_approximator.py,sha256=hX4nqa0yw1OkA2UKmzZ3HhvMfL0Wwv1yQePxrgAueS8,4782
+torchzero/modules/grad_approximation/rfdm.py,sha256=mYS89DI7Rel10QsB9Q2cyrjo4ChhPD1kX00h0t6dc4A,19599
+torchzero/modules/least_squares/__init__.py,sha256=mJwE2IXVB3mn_7BzsmDNKhfyViCV8GOrqHJJjz04HR4,41
+torchzero/modules/least_squares/gn.py,sha256=Yl6HgZuZdba83ytzi5B8zBIRE7KfDJiQeO3kxZVjmBM,7146
+torchzero/modules/line_search/__init__.py,sha256=_QjxUJmNC8OqtUuyTJp9wDfHNFKZBZqj6lttWKhG-cI,217
+torchzero/modules/line_search/_polyinterp.py,sha256=i3sNl6SFAUJi4oxhhjBlcxJY9KRunIZjJ8sGdaJOVjc,10990
+torchzero/modules/line_search/adaptive.py,sha256=YNabP6-01dhAUDAOuHRPZCwiV5xTRdHmkN667HQ6V3w,3798
+torchzero/modules/line_search/backtracking.py,sha256=laQAgy4a6JWovAbjpjGxFn323moU9r3jjv6I5R780sE,9047
+torchzero/modules/line_search/interpolation.py,sha256=tHXlZD1MgfYaymhXY75k9CocltwSYDYZg6ENDCEUiss,4942
+torchzero/modules/line_search/line_search.py,sha256=wqx65OkN3jkyuTuwCQtelVDDWbeWbPYV8Twi2jmrqaw,12926
+torchzero/modules/line_search/scipy.py,sha256=xQ80h9cSyF4Iorq_1NoJglu_Bx4_KeojulBIxvwU6gQ,2836
+torchzero/modules/line_search/strong_wolfe.py,sha256=KRvHnZ0q2nYFfLJry6-Iki936n7B9WwHYDqNgeBsIrs,14975
+torchzero/modules/misc/__init__.py,sha256=UYY9CeNepnC8H1LnFa829ux5MEjtGZ9zql624IbCFX8,825
+torchzero/modules/misc/debug.py,sha256=wFt9wB6IdRSsOGLhQjdjmGt4KdB0V5IT0iBFMj97R3Y,1617
+torchzero/modules/misc/escape.py,sha256=-7TTIZzxiJpVQ_jSFLy64LHijlJfC5y6THvdpNOfrWs,1928
+torchzero/modules/misc/gradient_accumulation.py,sha256=DKJOKewtdvKcSlQEU2ZifB9awdqXU5IxGv68L8yumb0,2269
+torchzero/modules/misc/homotopy.py,sha256=oa0YFYfv8kkg9v7nukdjTwinuyQa4Nt7kTpddUVCSKg,2257
+torchzero/modules/misc/misc.py,sha256=lnz071OhYCrpQ-QO8yU77ba3NLZEgvJwJ9dFUDhxupk,15000
+torchzero/modules/misc/multistep.py,sha256=Z0JL1PTRIwnLkBD0tAnxXjCsYJMAr3g3R79lz5DsHNQ,6779
+torchzero/modules/misc/regularization.py,sha256=P5sPHItQcuuOvLioA5x5Mk2ms33E2brjfMaWG0Yzlok,5931
+torchzero/modules/misc/split.py,sha256=PQvI5VPdQo1_Yr9ypn57ZBtkI3EiHovXvIjduoL-LBg,4286
+torchzero/modules/misc/switch.py,sha256=dboHoc1Ijj9ziQEx2HzbT4mnGQH6OVs673BfcfPzLls,3674
+torchzero/modules/momentum/__init__.py,sha256=AKWC4HIkN9ZJwN38dJvVJkFEhiP9r93G-kMDokBfsj8,281
+torchzero/modules/momentum/averaging.py,sha256=Q6WLwCJwgNY96YIfQXWpsX-2kDR7n0IOMDfZMvNVc9U,3035
+torchzero/modules/momentum/cautious.py,sha256=3xlWuuMAWrJGnvtbdNYZNB0xjDeo-gUaOfyCqxCdMGI,8373
+torchzero/modules/momentum/momentum.py,sha256=gxyOudXyABUBU4xodBvWm9Cr6SQgcwf5WEevQWhKjPA,4390
+torchzero/modules/ops/__init__.py,sha256=xUYzWWLlSwaT8sw3dWywkALqI6YGCZgptWQJVy83HhM,1249
+torchzero/modules/ops/accumulate.py,sha256=f-Uutg7gNFRobTc5YI9JlfFiSacXmg0gDhIwQNwZSZg,3439
+torchzero/modules/ops/binary.py,sha256=eB6zwz5ZSSyeWvwVfuOFMjem93oMB7hCo4kNF705jn8,12219
+torchzero/modules/ops/higher_level.py,sha256=F_fv12U1AioRSq0I1FNa9pXYWsyOj-albyTgU9XniiM,8966
+torchzero/modules/ops/multi.py,sha256=9TW9sDlhkewAIFk_YtIb_L9oppDCAh5LAvv3t7w77lM,8641
+torchzero/modules/ops/reduce.py,sha256=SzpkNV5NTsVFp-61a1m8lDKJ1ivJmfQofolFWxbbAe4,6526
+torchzero/modules/ops/unary.py,sha256=vXvWfDFo2CBFwb1ej_WV-fGg61lQRbwN4HklAik8tJY,4844
+torchzero/modules/ops/utility.py,sha256=UkR3BfN_NBsZe78oERnknf6lmgeGNiEbtfUDNgU0YoQ,4423
+torchzero/modules/projections/__init__.py,sha256=4LfmBEu_eM4YWmcWQVH4CdI1H0ucCIHDH9tTGigjVPY,136
+torchzero/modules/projections/cast.py,sha256=FJx2Tt1lbQRnOC5wxx3LbOnacLfUluFP6QOXLUCIEPY,2174
+torchzero/modules/projections/galore.py,sha256=70k30-2RJ3ncTNZpjxBhyYq1yJVFGfS2YUvU2trIK4o,258
+torchzero/modules/projections/projection.py,sha256=MMQOplkELu-3fG0qLyx0-uLcDQKCzp7EiCuqTEdPHZ0,14277
+torchzero/modules/quasi_newton/__init__.py,sha256=scn1qRpS1dtUv0u4tQPbjPy9Db66KrxyjS8YNDiKDYQ,543
+torchzero/modules/quasi_newton/damping.py,sha256=EQzxyRcq-KSQP9RJA8UscDo0LNigNXmdFBhyeiKN2p0,2807
+torchzero/modules/quasi_newton/diagonal_quasi_newton.py,sha256=Zx-tlFRa89GhoSP7RFJdLQJPiqPCL7rWaV7WJoQ1YCs,6930
+torchzero/modules/quasi_newton/lbfgs.py,sha256=tBmWXQ1AJE-yikRfx2g2xXRcTfIMzBEDCP_FPPpkMFM,11200
+torchzero/modules/quasi_newton/lsr1.py,sha256=EozLObU28uTg88ZwV4SqBeFrQbJXqK9TfZnKqYON7rk,8512
+torchzero/modules/quasi_newton/quasi_newton.py,sha256=B0712bqhrVNhtGiQAN6b4AgR__QdsVirLkMdOx1YGw4,45475
+torchzero/modules/quasi_newton/sg2.py,sha256=b9Ly-BN4q82oki6wRKvDitBUfoTr6u3aMIQhpBak62E,7836
+torchzero/modules/restarts/__init__.py,sha256=7282ePwN_I0vSeLPYS4TTclE9ZU7pL6UpyRp5ydgdSg,134
+torchzero/modules/restarts/restars.py,sha256=R-8eVQ1SqUhhZrr9CM2NcIlqs0HsFRaMwjXr6FyjD5w,9262
+torchzero/modules/second_order/__init__.py,sha256=42HeVA3Azl_tXV0_injU-q4QOu7lXzt6AVUcwnPy4Ag,313
+torchzero/modules/second_order/ifn.py,sha256=oAjfFVjLzG6L4n_ELXAWGZSicWizilQy_hQf4hmOoL0,2019
+torchzero/modules/second_order/inm.py,sha256=hafMvqQBwwhTpWElwWouOj_BuMR5v7NacZg-6dpPNbA,3226
+torchzero/modules/second_order/multipoint.py,sha256=mHG1SFLsILELIspxZ8U_hxJBlkGwzvUWg96bOIrQsIY,7500
+torchzero/modules/second_order/newton.py,sha256=kGAsZCYd9uMDLUg_dhrS58JsSVjnZ2eqnJmOr-fghRo,11588
+torchzero/modules/second_order/newton_cg.py,sha256=r9gb9hoxKwhQ1FIaZENZscnaPV_9fiJtmuhS7Dd-_7I,14810
+torchzero/modules/second_order/nystrom.py,sha256=GXff6kAge8v-Pt-rCgivEIqXa2xxoEPsByBC9Rkx0kM,11397
+torchzero/modules/second_order/rsn.py,sha256=uUvMXFh6tzciz92myBnsxF1rCGAY71oBPFC04mBGmV8,10369
+torchzero/modules/smoothing/__init__.py,sha256=RYxCLLfG2onBbMUToaoedsr20rXaayyBt7Ov8OxULrU,80
+torchzero/modules/smoothing/laplacian.py,sha256=Cs351WVfapQjK4loYlAT6Sx0xIHnc89dhOVNnxUJiGI,5199
+torchzero/modules/smoothing/sampling.py,sha256=tYWK0Rw1CRcultKbMK0C5-TUFXbffjR62338iYYtIck,12942
+torchzero/modules/step_size/__init__.py,sha256=jG0qXpIn17oYXL8b34UjiEbkl002hj3FqJk1uQ5bkCg,136
+torchzero/modules/step_size/adaptive.py,sha256=6YMnaHDtUgC3xdEx8dieDAKrEsYv51eP4craB4JpjP0,14600
+torchzero/modules/step_size/lr.py,sha256=O3LkxvmeUiDjmH6-oILh7WCfYcC8b44yiQDfVLL5TeU,5927
+torchzero/modules/termination/__init__.py,sha256=LkXBiOOYD4ce1Lemj0Vx9BCm_KhRTQTMvm-PD4lQwTs,344
+torchzero/modules/termination/termination.py,sha256=lJXLmtA84JoK_QHhjBxfW9lkxGIxqg7cE7d755MeduE,6905
+torchzero/modules/trust_region/__init__.py,sha256=kWke9FB41-EpjdXCPk8VBwZhpgYalOWSKDI1XWe0yYg,204
+torchzero/modules/trust_region/cubic_regularization.py,sha256=kz4mwo2cD6X6qQKVvjAA1PHbcZy3qLBiDZVKDpY7fxM,6709
+torchzero/modules/trust_region/dogleg.py,sha256=zwFR49gghxztVGEETF2D4AkeGgHkQRbHGGelav3GuFg,3619
+torchzero/modules/trust_region/levenberg_marquardt.py,sha256=lwjNyrvdOrigjXuToM9qQq4AhpRHQuJ9oYCGyuPcWEg,5035
+torchzero/modules/trust_region/trust_cg.py,sha256=mfTGsAZU0L46k2pdv3w663lXPCkNekbW2eTHQBED1iw,4453
+torchzero/modules/trust_region/trust_region.py,sha256=oXMNIvboz0R_1J0Gfd4IvbnwZFl32csNVv-lTYGB0zk,12913
+torchzero/modules/variance_reduction/__init__.py,sha256=3pwPWZpjgz1btfLJ3rEaK7Wl8B1pDh0HIf0kvD_NJH8,22
+torchzero/modules/variance_reduction/svrg.py,sha256=R3bULX3nw6M7hw4ZN0cpOvxvdy8CLPHYAStPY9qfDng,8804
+torchzero/modules/weight_decay/__init__.py,sha256=zQrjSujD0c-rKfKjUpuutfAODljsz1hS3zUNJW7zbh4,132
+torchzero/modules/weight_decay/reinit.py,sha256=ml70eC4FXh-9G_Iq8tqoJ9R6NBxetROIGp1YaqfGfMw,3326
+torchzero/modules/weight_decay/weight_decay.py,sha256=N9ozvYmEKFOpG84wWfgz7NifYQNwQTBwnXEtR7ygiN4,5352
+torchzero/modules/wrappers/__init__.py,sha256=6b5Ac-8u18IVp_Jnw1T1xQExwpQhpQ0JwNV9GyC_Yj8,31
+torchzero/modules/wrappers/optim_wrapper.py,sha256=P9Rsilde-SD6WwXQWDrJetZXw7NSowZznvCuFQkps8s,4693
+torchzero/modules/zeroth_order/__init__.py,sha256=1ADUiOHVHzvIP4TpH7_ILmeW2heidfikbf6d5g_1RzY,18
+torchzero/modules/zeroth_order/cd.py,sha256=fTLDtrX4nLX1vG7_isxIUtAkYnxCKS_cCECnPyHTQq0,5017
+torchzero/optim/__init__.py,sha256=aXf7EkywqYiR50I4QeeVXro9aBhKiqfbY_BCia59sgU,46
+torchzero/optim/root.py,sha256=MnXytsgTjDKhYw3UKux1O-g4vuHsFQu2Emsq5zphu_8,2308
+torchzero/optim/utility/__init__.py,sha256=pUacok4XmebfxofE-QWZLgViajsU-3JkXcWi9OS-Jrw,24
+torchzero/optim/utility/split.py,sha256=uNnKA40OiXUlr-vlHuU_rLEUfXQXgvr6Cd9yGzjWJiA,1702
+torchzero/optim/wrappers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+torchzero/optim/wrappers/directsearch.py,sha256=7f1Zy9ZxftgIFxJ9_1snyMjp8ypB4ULyddOqgxdwkEg,9806
+torchzero/optim/wrappers/fcmaes.py,sha256=Bobx87TWinICd3SyNaKhoplpNbrLdD5vR1J1TZw69Cs,3437
+torchzero/optim/wrappers/mads.py,sha256=Gm1msCsVNJDq-WS913peZ0TzXU_7wgP_sPmMfEK0VIY,2326
+torchzero/optim/wrappers/moors.py,sha256=AjEeovPEvxEGJwbhOaOCrCIxPgwOHilkj-vWb-88Y_0,2073
+torchzero/optim/wrappers/nevergrad.py,sha256=Tmo-9RaFH7OXTJU36UMSALJro2K0AuQk6ow9g9ZoPoU,4890
+torchzero/optim/wrappers/nlopt.py,sha256=FzU_3VuNN0_ngT_FlQs7ZOL_RZcfeFtJNmklqvyzTkw,8694
+torchzero/optim/wrappers/optuna.py,sha256=31K-oeffoKfBc-LmGY5F33g41qlK-RDjZsZO-mZbSRs,2211
+torchzero/optim/wrappers/pybobyqa.py,sha256=1eH3HW2HVyNneZ5vB9ch1hb0OdoYrnk4pJmQVRIOAw4,6309
+torchzero/optim/wrappers/wrapper.py,sha256=eEZnqKRc2cSFVVXNyOGGReMX3MvY0eoHSRjFsQ8pCSw,4380
+torchzero/optim/wrappers/scipy/__init__.py,sha256=lAciW9WTZ8JZ9ZFunyi78cOa6URhTL-rIjk1AdqbrxM,262
+torchzero/optim/wrappers/scipy/basin_hopping.py,sha256=11FuVLYDJxbyxW-44UO-JRef7MErGFfM_PqqqADA8RQ,4435
+torchzero/optim/wrappers/scipy/brute.py,sha256=MD5fWcUo0-COrq5_OKM8X5SSMZaN--4ThW8mdISLX00,1213
+torchzero/optim/wrappers/scipy/differential_evolution.py,sha256=GBJSh0K5ueZX3fY9vxTntQnuIoLDjOQyOj6TMCmTGBg,2669
+torchzero/optim/wrappers/scipy/direct.py,sha256=lTkH8aRY7sWkhrtQ8Z-m8fSdhTpsLxOIQSJprSqThFA,1821
+torchzero/optim/wrappers/scipy/dual_annealing.py,sha256=mIdXy8F3YEfrQCXNxps8T33yRDuqJxaaIjzB0wlKo90,4117
+torchzero/optim/wrappers/scipy/experimental.py,sha256=-_82LtrDVg0qRg6E1coXugsCGGcVNXgxv6m5Anaaw24,4833
+torchzero/optim/wrappers/scipy/minimize.py,sha256=6O2Js2ecGZkNE3mqXJR9Rqp5xlh3KIVYBMF82k5hrws,6491
+torchzero/optim/wrappers/scipy/sgho.py,sha256=mKG9uG-zaEzB6mbiZ25FnXdRcfvGYYOHapuHniyJ8r8,4054
+torchzero/utils/__init__.py,sha256=tpy9ti5Ub5d1zQPFODZ4PjmFKNzoZTd-NByd_snlYtk,761
+torchzero/utils/compile.py,sha256=dY9ioWQvJt71HQN_z4kXX4KgtJ0xOW0MKNlZpHD6118,5130
+torchzero/utils/derivatives.py,sha256=Crma5mVdI4hTPMkoMzEfMTAyCXZIUuzeEzkzQMpBNOY,18089
+torchzero/utils/metrics.py,sha256=XPpOvY257tb4mN3Sje1AVNlQkOXiW24_lXXdtd0JYok,3130
+torchzero/utils/numberlist.py,sha256=iMoqz4IzXy-aE9bqVYJ21GV6pl0z-NeTsXR-LaI8C24,6229
+torchzero/utils/optimizer.py,sha256=G741IvE57RaVYowr9FEqfRm_opPAeu4UWKU5iPKDMFA,8415
+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=HATghTNijlQxmw8rzJfZPPGj1CjcnRxEwogmrgqnARU,4577
+torchzero/utils/tensorlist.py,sha256=4rN8gm967pPmtO5kotXqIX7Mal0ps-IHkGBybfeWY4M,56357
+torchzero/utils/thoad_tools.py,sha256=G8k-z0vireEUtI3A_YAR6dtwYjSnN49e_GadcHwwQKc,2319
+torchzero/utils/torch_tools.py,sha256=DsHaSRGZ3-IuySZJTrkojTbaMMlttJFe0hFvB2xnl2U,5069
+torchzero-0.4.0.dist-info/METADATA,sha256=WbRYUNngrfEBuPLquDDkrl31aqVDq4mTacBurEfdgmI,564
+torchzero-0.4.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+torchzero-0.4.0.dist-info/top_level.txt,sha256=ETW_iE2ubg0oMyef_h-ayB5i1OOZZd4SNdR3ltIbHe0,16
+torchzero-0.4.0.dist-info/RECORD,,

tests/test_vars.py

@@ -1,185 +0,0 @@
-import pytest
-import torch
-from torchzero.core.module import Var
-from torchzero.utils.tensorlist import TensorList
-
-@torch.no_grad
-def test_var_get_loss():
-
-    # ---------------------------- test that it works ---------------------------- #
-    params = [torch.tensor(2.0, requires_grad=True)]
-    evaluated = False
-
-    def closure_1(backward=True):
-        assert not backward, 'backward = True'
-
-        # ensure closure only evaluates once
-        nonlocal evaluated
-        assert evaluated is False, 'closure was evaluated twice'
-        evaluated = True
-
-        loss = params[0]**2
-        if backward:
-            params[0].grad = None
-            loss.backward()
-        else:
-            assert not loss.requires_grad, "loss requires grad with backward=False"
-        return loss
-
-    var = Var(params=params, closure=closure_1, model=None, current_step=0)
-
-    assert var.loss is None, var.loss
-
-    assert (loss := var.get_loss(backward=False)) == 4.0, loss
-    assert evaluated, evaluated
-    assert loss is var.loss
-    assert var.loss == 4.0
-    assert var.loss_approx == 4.0
-    assert var.grad is None, var.grad
-
-    # reevaluate, which should just return already evaluated loss
-    assert (loss := var.get_loss(backward=False)) == 4.0, loss
-    assert var.grad is None, var.grad
-
-
-    # ----------------------- test that backward=True works ---------------------- #
-    params = [torch.tensor(3.0, requires_grad=True)]
-    evaluated = False
-
-    def closure_2(backward=True):
-        # ensure closure only evaluates once
-        nonlocal evaluated
-        assert evaluated is False, 'closure was evaluated twice'
-        evaluated = True
-
-        loss = params[0] * 2
-        if backward:
-            assert loss.requires_grad, "loss does not require grad so `with torch.enable_grad()` context didn't work"
-            params[0].grad = None
-            loss.backward()
-        else:
-            assert not loss.requires_grad, "loss requires grad with backward=False"
-        return loss
-
-    var = Var(params=params, closure=closure_2, model=None, current_step=0)
-    assert var.grad is None, var.grad
-    assert (loss := var.get_loss(backward=True)) == 6.0, loss
-    assert var.grad is not None
-    assert var.grad[0] == 2.0, var.grad
-
-    # reevaluate, which should just return already evaluated loss
-    assert (loss := var.get_loss(backward=True)) == 6.0, loss
-    assert var.grad[0] == 2.0, var.grad
-
-    # get grad, which should just return already evaluated grad
-    assert (grad := var.get_grad())[0] == 2.0, grad
-    assert grad is var.grad, grad
-
-    # get update, which should create and return cloned grad
-    assert var.update is None
-    assert (update := var.get_update())[0] == 2.0, update
-    assert update is var.update
-    assert update is not var.grad
-    assert var.grad is not None
-    assert update[0] == var.grad[0]
-
-@torch.no_grad
-def test_var_get_grad():
-    params = [torch.tensor(2.0, requires_grad=True)]
-    evaluated = False
-
-    def closure(backward=True):
-        # ensure closure only evaluates once
-        nonlocal evaluated
-        assert evaluated is False, 'closure was evaluated twice'
-        evaluated = True
-
-        loss = params[0]**2
-        if backward:
-            assert loss.requires_grad, "loss does not require grad so `with torch.enable_grad()` context didn't work"
-            params[0].grad = None
-            loss.backward()
-        else:
-            assert not loss.requires_grad, "loss requires grad with backward=False"
-        return loss
-
-    var = Var(params=params, closure=closure, model=None, current_step=0)
-    assert (grad := var.get_grad())[0] == 4.0, grad
-    assert grad is var.grad
-
-    assert var.loss == 4.0
-    assert (loss := var.get_loss(backward=False)) == 4.0, loss
-    assert (loss := var.get_loss(backward=True)) == 4.0, loss
-    assert var.loss_approx == 4.0
-
-    assert var.update is None, var.update
-    assert (update := var.get_update())[0] == 4.0, update
-
-@torch.no_grad
-def test_var_get_update():
-    params = [torch.tensor(2.0, requires_grad=True)]
-    evaluated = False
-
-    def closure(backward=True):
-        # ensure closure only evaluates once
-        nonlocal evaluated
-        assert evaluated is False, 'closure was evaluated twice'
-        evaluated = True
-
-        loss = params[0]**2
-        if backward:
-            assert loss.requires_grad, "loss does not require grad so `with torch.enable_grad()` context didn't work"
-            params[0].grad = None
-            loss.backward()
-        else:
-            assert not loss.requires_grad, "loss requires grad with backward=False"
-        return loss
-
-    var = Var(params=params, closure=closure, model=None, current_step=0)
-    assert var.update is None, var.update
-    assert (update := var.get_update())[0] == 4.0, update
-    assert update is var.update
-
-    assert (grad := var.get_grad())[0] == 4.0, grad
-    assert grad is var.grad
-    assert grad is not update
-
-    assert var.loss == 4.0
-    assert (loss := var.get_loss(backward=False)) == 4.0, loss
-    assert (loss := var.get_loss(backward=True)) == 4.0, loss
-    assert var.loss_approx == 4.0
-
-    assert (update := var.get_update())[0] == 4.0, update
-
-
-def _assert_var_are_same_(v1: Var, v2: Var, clone_update: bool):
-    for k,v in v1.__dict__.items():
-        if not k.startswith('__'):
-            # if k == 'post_step_hooks': continue
-            if k == 'storage': continue
-            if k == 'update' and clone_update:
-                if v1.update is None or v2.update is None:
-                    assert v1.update is None and v2.update is None, f'{k} is not the same, {v1 = }, {v2 = }'
-                else:
-                    assert (TensorList(v1.update) == TensorList(v2.update)).global_all()
-                    assert v1.update is not v2.update
-            else:
-                assert getattr(v2, k) is v, f'{k} is not the same, {v1 = }, {v2 = }'
-
-def test_var_clone():
-    model = torch.nn.Sequential(torch.nn.Linear(2,2), torch.nn.Linear(2,4))
-    def closure(backward): return 1
-    var = Var(params=list(model.parameters()), closure=closure, model=model, current_step=0)
-
-    _assert_var_are_same_(var, var.clone(clone_update=False), clone_update=False)
-    _assert_var_are_same_(var, var.clone(clone_update=True), clone_update=True)
-
-    var.grad = TensorList(torch.randn(5))
-    _assert_var_are_same_(var, var.clone(clone_update=False), clone_update=False)
-    _assert_var_are_same_(var, var.clone(clone_update=True), clone_update=True)
-
-    var.update = TensorList(torch.randn(5) * 2)
-    var.loss = torch.randn(1)
-    var.loss_approx = var.loss
-    _assert_var_are_same_(var, var.clone(clone_update=False), clone_update=False)
-    _assert_var_are_same_(var, var.clone(clone_update=True), clone_update=True)

torchzero/modules/experimental/momentum.py

@@ -1,160 +0,0 @@
-from collections.abc import Sequence
-from functools import partial
-from operator import itemgetter
-from typing import Literal
-
-import torch
-
-from ...core import Target, Transform
-from ...utils import NumberList, TensorList, unpack_dicts, unpack_states
-from ..functional import ema_, ema_sq_, sqrt_ema_sq_
-from ..momentum.momentum import nag_
-from ..ops.higher_level import EMASquared, SqrtEMASquared
-
-
-def precentered_ema_sq_(
-    tensors: TensorList,
-    exp_avg_: TensorList,
-    exp_avg_sq_: TensorList,
-    beta1: float | NumberList,
-    beta2: float | NumberList,
-    step: int,
-    min_step: int,
-    pow: float,
-    max_exp_avg_sq_: TensorList | None,
-):
-    """
-    Squared EMA of (update - 1st EMA). Starts taking effect after `min_step` to avoid division by epsilon.
-
-    returns `exp_avg_sq_` or `max_exp_avg_sq_`.
-    """
-    exp_avg_ = ema_(tensors, exp_avg_=exp_avg_, beta=beta1, dampening=0, lerp=False)
-
-    if step < min_step: centered_update = tensors
-    else: centered_update = tensors - exp_avg_
-
-    exp_avg_sq_=ema_sq_(
-        centered_update,
-        exp_avg_sq_=exp_avg_sq_,
-        beta=beta2,
-        pow=pow,
-        max_exp_avg_sq_=max_exp_avg_sq_,
-    )
-    return exp_avg_sq_
-
-class PrecenteredEMASquared(Transform):
-    """Maintains un-squared EMA, the updates are centered by it before being fed into squared EMA."""
-    def __init__(self, beta1:float=0.99, beta2=0.99, min_step: int = 2, amsgrad=False, pow:float=2, target: Target = 'update'):
-        defaults = dict(beta1=beta1,beta2=beta2,pow=pow,amsgrad=amsgrad, min_step=min_step)
-        super().__init__(defaults, uses_grad=False, target=target)
-
-    @torch.no_grad
-    def apply_tensors(self, tensors, params, grads, loss, states, settings):
-        step = self.global_state['step'] = self.global_state.get('step', 0) + 1
-
-        beta1, beta2 = unpack_dicts(settings, 'beta1','beta2', cls=NumberList)
-        amsgrad, pow, min_step = itemgetter('amsgrad', 'pow', 'min_step')(settings[0])
-
-        if amsgrad:
-            exp_avg, exp_avg_sq, max_exp_avg_sq = unpack_states(states, tensors, 'exp_avg', 'exp_avg_sq', 'max_exp_avg_sq', cls=TensorList)
-        else:
-            exp_avg, exp_avg_sq = unpack_states(states, tensors, 'exp_avg', 'exp_avg_sq', cls=TensorList)
-            max_exp_avg_sq = None
-
-        return precentered_ema_sq_(
-            TensorList(tensors),
-            exp_avg_ = exp_avg,
-            exp_avg_sq_=exp_avg_sq,
-            beta1=beta1,
-            beta2=beta2,
-            step = step,
-            min_step=min_step,
-            pow=pow,
-            max_exp_avg_sq_=max_exp_avg_sq,
-        ).clone()
-
-
-def nag_ema_sq_(
-    tensors: TensorList,
-    exp_avg_sq_: TensorList,
-    beta: float | NumberList,
-    max_exp_avg_sq_: TensorList | None,
-    pow: float,
-    lerp:bool=True,
-):
-    """
-    Nesterov EMA of squared tensors.
-
-    Returns `exp_avg_sq_` or `max_exp_avg_sq_`.
-    """
-    if pow == 1: tensors = tensors.abs()
-    elif pow%2 == 0: tensors = tensors.pow(pow)
-    else: tensors = tensors.pow(pow).abs()
-
-    exp_avg_sq_=nag_(tensors,velocity_=exp_avg_sq_,momentum=beta,dampening=0,lerp=lerp,)
-
-    # AMSGrad
-    if max_exp_avg_sq_ is not None:
-        max_exp_avg_sq_.maximum_(exp_avg_sq_)
-        exp_avg_sq_ = max_exp_avg_sq_
-
-    return exp_avg_sq_
-
-def sqrt_nag_ema_sq_(
-    tensors: TensorList,
-    exp_avg_sq_: TensorList,
-    beta: float | NumberList,
-    max_exp_avg_sq_: TensorList | None,
-    debiased: bool,
-    step: int,
-    pow: float,
-    lerp:bool=False,
-):
-    """
-    Square root of nesterov EMA of squared tensors.
-
-    Returns new tensors.
-    """
-    return sqrt_ema_sq_(tensors=tensors,exp_avg_sq_=exp_avg_sq_,beta=beta,max_exp_avg_sq_=max_exp_avg_sq_,
-                        pow=pow,debiased=debiased,step=step,ema_sq_fn=partial(nag_ema_sq_,lerp=lerp))
-
-class NesterovEMASquared(EMASquared):
-    """squared momentum with nesterov momentum rule"""
-    EMA_SQ_FN = staticmethod(nag_ema_sq_)
-
-class SqrtNesterovEMASquared(SqrtEMASquared):
-    """square root of squared momentum with nesterov momentum rule"""
-    SQRT_EMA_SQ_FN = staticmethod(sqrt_nag_ema_sq_)
-
-
-def coordinate_momentum_(
-    tensors: TensorList,
-    velocity_: TensorList,
-    p: float | NumberList,
-):
-    """
-    sets `velocity_` to p% random values from `tensors`.
-
-    Returns `velocity_`
-    """
-    mask = tensors.bernoulli_like(p).as_bool()
-    velocity_.masked_set_(mask, tensors)
-    return velocity_
-
-
-class CoordinateMomentum(Transform):
-    """Maintains a momentum buffer, on each step each value in the buffer has :code:`p` chance to be updated with the new value.
-
-    Args:
-        p (float, optional): _description_. Defaults to 0.1.
-        target (Target, optional): _description_. Defaults to 'update'.
-    """
-    def __init__(self, p: float = 0.1, target: Target = 'update'):
-        defaults = dict(p=p)
-        super().__init__(defaults, uses_grad=False, target=target)
-
-    @torch.no_grad
-    def apply_tensors(self, tensors, params, grads, loss, states, settings):
-        p = NumberList(s['p'] for s in settings)
-        velocity = unpack_states(states, tensors, 'velocity', cls=TensorList)
-        return coordinate_momentum_(TensorList(tensors), velocity_=velocity, p=p).clone()

torchzero/modules/higher_order/__init__.py

@@ -1 +0,0 @@
-from .higher_order_newton import HigherOrderNewton