python-somax 0.0.1__py3-none-any.whl

python_somax-0.0.1.dist-info/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

python_somax-0.0.1.dist-info/METADATA

@@ -0,0 +1,131 @@
+Metadata-Version: 2.1
+Name: python-somax
+Version: 0.0.1
+Summary: SOMAX: Second-Order Methods for Machine Learning in JAX
+Author: Nick Korbit
+Platform: any
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: jaxopt >=0.8.2
+
+<h1 align='center'>Somax</h1>
+
+<p align='center'>
+  <img src="assets/somax_logo_mini.png" alt="logo" width="250px"/>
+</p>
+
+Somax is a library of Second-Order Methods for stochastic optimization
+written in [JAX](https://github.com/google/jax).
+Somax is based on the [JAXopt](https://github.com/google/jaxopt) StochasticSolver API,
+and can be used as a drop-in
+replacement for JAXopt as well as
+[Optax](https://github.com/google-deepmind/optax) solvers.
+
+Currently supported methods:
+
+- Diagonal Scaling:
+    - [AdaHessian](https://ojs.aaai.org/index.php/AAAI/article/view/17275);
+    - [Sophia](https://arxiv.org/abs/2305.14342);
+- Hessian-free Optimization:
+    - [Newton-CG](https://epubs.siam.org/doi/10.1137/10079923X);
+- Quasi-Newton:
+    - [Stochastic Quasi-Newton Framework (SQN)](https://arxiv.org/abs/1606.04838);
+- Gauss-Newton:
+    - [Exact Gauss-Newton (EGN)](https://arxiv.org/abs/2405.14402);
+    - [Stochastic Gauss-Newton (SGN)](https://arxiv.org/abs/2006.02409);
+- Natural Gradient:
+    - [Natural Gradient with Sherman-Morrison-Woodbury formula (SWM-NG)](https://arxiv.org/abs/1906.02353).
+
+Future releases:
+
+- Add support for separate "gradient batches"
+  and "curvature batches" for all solvers;
+- Add support for Optax rate schedules.
+
+⚠️ Since JAXopt is currently being merged into Optax,
+Somax at some point will switch to the Optax API as well.
+
+
+*The catfish in the logo is a nod to "сом", 
+the Belarusian word for "catfish", also pronounced as "som".
+
+
+## Installation
+
+```bash
+pip install python-somax
+```
+
+Requires [JAXopt](https://github.com/google/jaxopt) 0.8.2+.
+
+## Quick example
+
+```py
+from somax import EGN
+
+# initialize the solver
+solver = EGN(
+    predict_fun=model.apply,
+    loss_type='mse',
+    learning_rate=0.1,
+    regularizer=1.0,
+)
+
+# initialize the solver state
+opt_state = solver.init_state(params)
+
+# run the optimization loop
+for i in range(10):
+    params, opt_state = solver.update(params, opt_state, batch_x, batch_y)
+```
+
+See more in the [examples](examples) folder.
+
+## Citation
+
+```bibtex
+@misc{korbit2024somax,
+  author = {Nick Korbit},
+  title = {{SOMAX}: a library of second-order methods for stochastic optimization written in {JAX}},
+  year = {2024},
+  url = {https://github.com/cor3bit/somax},
+}
+```
+
+## See also
+
+**Optimization with JAX**  
+[Optax](https://github.com/google-deepmind/optax): first-order gradient (SGD, Adam, ...) optimisers.  
+[JAXopt](https://github.com/google/jaxopt): deterministic second-order methods (e.g., Gauss-Newton, Levenberg
+Marquardt), stochastic first-order methods PolyakSGD, ArmijoSGD.
+
+**Awesome Projects**  
+[Awesome JAX](https://github.com/n2cholas/awesome-jax): a longer list of various JAX projects.  
+[Awesome SOMs](https://github.com/cor3bit/awesome-soms): a list
+of resources for second-order optimization methods in machine learning.
+
+## Acknowledgements
+
+Some of the implementation ideas are based on the following repositories:
+
+- Line Search in JAXopt: https://github.com/google/jaxopt/blob/main/jaxopt/_src/armijo_sgd.py#L48
+
+- L-BFGS Inverse Hessian-Gradient product in JAXopt: https://github.com/google/jaxopt/blob/main/jaxopt/_src/lbfgs.py#L44
+
+- AdaHessian (official implementation): https://github.com/amirgholami/adahessian
+
+- AdaHessian (Nestor Demeure's implementation): https://github.com/nestordemeure/AdaHessianJax
+
+- Sophia (official implementation): https://github.com/Liuhong99/Sophia
+
+- Sophia (levanter implementation): https://github.com/stanford-crfm/levanter/blob/main/src/levanter/optim/sophia.py
+

python_somax-0.0.1.dist-info/RECORD

@@ -0,0 +1,19 @@
+somax/__init__.py,sha256=aAvLcWAeV1eUf5eQEhQkOd7VtQ7mQ1zIGnnzz4hffE8,445
+somax/diagonal/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+somax/diagonal/adahessian.py,sha256=iYqRiayhu3x_cSiJ_zaWaVTT6Avi2cFQD9qC-LWbp2U,6530
+somax/diagonal/sophia_g.py,sha256=c0kMYU8mHU8IXUKkLp3-PeP71J-P0XQJ7zzH0RuzGkw,6670
+somax/diagonal/sophia_h.py,sha256=eLsyDc9pCySyEZxugu0R4hFTraH_nZmyrnJJsdtxSO4,6096
+somax/gn/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+somax/gn/egn.py,sha256=qykmnvopfQdp-neOuyTU1DuAEKVqcno-MpaQPRUetlA,20880
+somax/gn/sgn.py,sha256=EzoJgn0nEuo6wNbnA_uacKHlzvbc3dYLe7vUtccOb7Y,7693
+somax/hf/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+somax/hf/newton_cg.py,sha256=lcvgJ9s6bHJwggKPBsRYvOYlyWxoShjV5IhdTYdYjlM,11735
+somax/ng/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+somax/ng/swm_ng.py,sha256=UFagbtVU16otMgw-_8_G238omSV7zQao2Ygg9OALgnk,14287
+somax/qn/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+somax/qn/sqn.py,sha256=6oKsIcZEpXtLQt1Wy2AquyIc_iy1SrBarivawO6PWVo,10261
+python_somax-0.0.1.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+python_somax-0.0.1.dist-info/METADATA,sha256=SL3R5GJsrQ1GxMb1DFqUEk4MSYEkLVr7QJ7oJ-BVtS8,4237
+python_somax-0.0.1.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
+python_somax-0.0.1.dist-info/top_level.txt,sha256=MtQ05b9WUJTb2l52XYjcvmLKK3LwGORMIW5Bl1mVUq0,6
+python_somax-0.0.1.dist-info/RECORD,,

python_somax-0.0.1.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: bdist_wheel (0.43.0)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

python_somax-0.0.1.dist-info/top_level.txt

@@ -0,0 +1 @@
+somax

somax/__init__.py

@@ -0,0 +1,19 @@
+# Diagonal Scaling
+from somax.diagonal.adahessian import AdaHessian
+from somax.diagonal.sophia_g import SophiaG
+from somax.diagonal.sophia_h import SophiaH
+
+# Hessian Free Optimization
+from somax.hf.newton_cg import NewtonCG
+
+# Quasi-Newton Methods
+from somax.qn.sqn import SQN
+
+# Gauss-Newton Methods
+from somax.gn.egn import EGN
+from somax.gn.sgn import SGN
+
+# Natural Gradient Methods
+from somax.ng.swm_ng import SWMNG
+
+__version__ = '0.0.1'

somax/diagonal/adahessian.py

@@ -0,0 +1,192 @@
+"""
+AdaHessian Optimizer
+Paper: https://arxiv.org/abs/2006.00719
+"""
+
+from typing import Any
+from typing import Callable
+from typing import NamedTuple, Tuple
+from typing import Optional
+import dataclasses
+from functools import partial
+
+import jax
+import jax.numpy as jnp
+from jax.flatten_util import ravel_pytree
+from jaxopt._src import base
+
+
+class AdaHessianState(NamedTuple):
+    """Named tuple containing state information."""
+    iter_num: int
+    stepsize: float
+    velocity_m: Any
+    velocity_v_tree: Any
+    hess_approx_rng: Any
+
+
+@dataclasses.dataclass(eq=False)
+class AdaHessian(base.StochasticSolver):
+    loss_fun: Callable
+
+    learning_rate: float = 1e-3
+
+    spatial_averaging: bool = True
+
+    # Momentum parameters
+    beta1: float = 0.9  # the exponential decay rate for the first moment estimates
+    beta2: float = 0.999  # the exponential decay rate for the squared hessian estimates
+
+    # Regularization parameters
+    weight_decay: float = 0.0  # L2 regularization coefficient
+
+    # Hessian parameters
+    hessian_power: float = 1.0  # the power of the hessian approximation, i.e. H^{-hessian_power}
+    eps: float = 1e-8  # term added to the denominator to improve numerical stability
+
+    hess_approx_seed: int = 1337
+
+    pre_update: Optional[Callable] = None
+
+    verbose: int = 0
+
+    jit: bool = True
+    unroll: base.AutoOrBoolean = "auto"
+
+    def __post_init__(self):
+        super().__post_init__()
+
+        self.reference_signature = self.loss_fun
+
+        self.grad_fun = jax.grad(self.loss_fun)
+
+        assert 0 <= self.hessian_power <= 1, "Hessian power must be in [0, 1]"
+        assert 0 <= self.weight_decay < 1, "Weight decay must be in [0, 1)"
+
+    def update(
+            self,
+            params: Any,
+            state: AdaHessianState,
+            *args,
+            **kwargs,
+    ) -> base.OptStep:
+        """Performs one iteration of the solver.
+
+        Args:
+          params: pytree containing the parameters.
+          state: named tuple containing the solver state.
+          *args: additional positional arguments to be passed to ``fun``.
+          **kwargs: additional keyword arguments to be passed to ``fun``.
+        Returns:
+          (params, state)
+        """
+        # ------- Step 1 -------
+        # compute the gradient and the Diagonal Hessian using Hutchinson's method
+        next_rng_key, rng_key = jax.random.split(state.hess_approx_rng)
+        params_flat, pack_fn = ravel_pytree(params)
+        p_shape = params_flat.shape
+        grads_tree, diag_hess_tree = self.grad_and_hess(params, rng_key, p_shape, pack_fn, *args, **kwargs)
+
+        # ------- Step 2 -------
+        if self.spatial_averaging:
+            diag_hess_tree = self.average_layers(diag_hess_tree)
+
+        # ------- Step 3 -------
+        # apply temporal averaging (momentum) to the first and second moments
+        i = state.iter_num + 1
+
+        # m_t
+        grads = ravel_pytree(grads_tree)[0]
+        velocity_m = self.beta1 * state.velocity_m + (1 - self.beta1) * grads
+        bias_corr_m = velocity_m / (1 - self.beta1 ** i)
+
+        # v_t
+        velocity_v_tree = jax.tree_map(lambda v, h: self.beta2 * v + (1 - self.beta2) * h,
+                                       state.velocity_v_tree, diag_hess_tree)
+
+        velocity_v = ravel_pytree(velocity_v_tree)[0]
+
+        bias_corr_v = jnp.pow(jnp.sqrt(velocity_v / (1 - self.beta2 ** i)), self.hessian_power)
+
+        # m_t / (v_t + eps)
+        direction = -bias_corr_m / (bias_corr_v + self.eps)
+
+        # update parameters
+        next_params_flat = params_flat + self.learning_rate * direction
+
+        # ------- Step 4 -------
+        # AdamW-style weight decay
+        if self.weight_decay > 0:
+            next_params_flat -= self.learning_rate * self.weight_decay * params_flat
+
+        # bookkeeping
+        next_params = pack_fn(next_params_flat)
+
+        next_state = AdaHessianState(
+            iter_num=state.iter_num + 1,
+            stepsize=state.stepsize,
+            velocity_m=velocity_m,
+            velocity_v_tree=velocity_v_tree,
+            hess_approx_rng=next_rng_key,
+        )
+
+        return base.OptStep(params=next_params, state=next_state)
+
+    def init_state(self,
+                   init_params: Any,
+                   *args,
+                   **kwargs) -> AdaHessianState:
+        params_flat, pack_fn = ravel_pytree(init_params)
+        velocity_m = jnp.zeros_like(params_flat)
+        velocity_v_tree = pack_fn(jnp.zeros_like(params_flat))
+
+        return AdaHessianState(
+            iter_num=0,
+            stepsize=self.learning_rate,
+            velocity_m=velocity_m,
+            velocity_v_tree=velocity_v_tree,
+            hess_approx_rng=jax.random.PRNGKey(self.hess_approx_seed),
+        )
+
+    def optimality_fun(self, params, *args, **kwargs):
+        """Optimality function mapping compatible with ``@custom_root``."""
+        # return self._grad_fun(params, *args, **kwargs)[0]
+        raise NotImplementedError
+
+    def __hash__(self):
+        # We assume that the attribute values completely determine the solver.
+        return hash(self.attribute_values())
+
+    def grad_and_hess(self, params, rng_key, p_shape, pack_fn, *args, **kwargs):
+        # prep grad function to accept only "params"
+        def inner_grad_fun(params):
+            return self.grad_fun(params, *args, **kwargs)
+
+        # draw from Radamacher distribution
+        z = jax.random.rademacher(rng_key, shape=p_shape, dtype=jnp.float32)
+        z_tree = pack_fn(z)
+
+        grads_tree, hz_tree = jax.jvp(inner_grad_fun, (params,), (z_tree,))
+
+        diag_hess_tree = jax.tree_map(lambda x, y: x * y, grads_tree, hz_tree)
+
+        return grads_tree, diag_hess_tree
+
+    @staticmethod
+    def _average_layer(leaf_node):
+        #  !! no abs() in the original implementation
+        # see also https://github.com/nestordemeure/AdaHessianJax/blob/main/adahessianJax/hessian_computation.py
+        abs_leaf_node = jnp.abs(leaf_node)
+
+        # !! no block size, just average over the layer
+        if leaf_node.ndim == 1:
+            return abs_leaf_node
+        elif leaf_node.ndim in [2, 3]:  # Convolution layers
+            return jnp.mean(abs_leaf_node, axis=-1, keepdims=True)
+        elif leaf_node.ndim == 4:  # Attention layers
+            return jnp.mean(abs_leaf_node, axis=[-2, -1], keepdims=True)
+        else:
+            raise ValueError("Invalid leaf node dimension")
+
+    def average_layers(self, diag_hess_tree):
+        return jax.tree_map(self._average_layer, diag_hess_tree)