hippoformer 0.0.12__tar.gz

hippoformer-0.0.12/.github/workflows/python-publish.yml

@@ -0,0 +1,36 @@
+# This workflow will upload a Python Package using Twine when a release is created
+# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries
+
+# This workflow uses actions that are not certified by GitHub.
+# They are provided by a third-party and are governed by
+# separate terms of service, privacy policy, and support
+# documentation.
+
+name: Upload Python Package
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  deploy:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python
+      uses: actions/setup-python@v2
+      with:
+        python-version: '3.x'
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        pip install build
+    - name: Build package
+      run: python -m build
+    - name: Publish package
+      uses: pypa/gh-action-pypi-publish@27b31702a0e7fc50959f5ad993c78deac1bdfc29
+      with:
+        user: __token__
+        password: ${{ secrets.PYPI_API_TOKEN }}

hippoformer-0.0.12/.github/workflows/test.yml

@@ -0,0 +1,21 @@
+name: Pytest
+on: [push, pull_request]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v4
+    - name: Set up Python 3.10
+      uses: actions/setup-python@v5
+      with:
+        python-version: "3.10"
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install -e .[test]
+    - name: Test with pytest
+      run: |
+        python -m pytest tests/

hippoformer-0.0.12/.gitignore

@@ -0,0 +1,207 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+#poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+#pdm.lock
+#pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+#pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Abstra
+# Abstra is an AI-powered process automation framework.
+# Ignore directories containing user credentials, local state, and settings.
+# Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#  Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#  that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#  and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#  you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Cursor
+#  Cursor is an AI-powered code editor. `.cursorignore` specifies files/directories to
+#  exclude from AI features like autocomplete and code analysis. Recommended for sensitive data
+#  refer to https://docs.cursor.com/context/ignore-files
+.cursorignore
+.cursorindexingignore
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/

hippoformer-0.0.12/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Phil Wang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

hippoformer-0.0.12/PKG-INFO

@@ -0,0 +1,65 @@
+Metadata-Version: 2.4
+Name: hippoformer
+Version: 0.0.12
+Summary: hippoformer
+Project-URL: Homepage, https://pypi.org/project/hippoformer/
+Project-URL: Repository, https://github.com/lucidrains/hippoformer
+Author-email: Phil Wang <lucidrains@gmail.com>
+License: MIT License
+        
+        Copyright (c) 2025 Phil Wang
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Keywords: artificial intelligence,deep learning,hippocampus,memory
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Requires-Dist: assoc-scan
+Requires-Dist: beartype
+Requires-Dist: einops>=0.8.1
+Requires-Dist: einx>=0.3.0
+Requires-Dist: torch>=2.4
+Requires-Dist: x-mlps-pytorch
+Provides-Extra: examples
+Provides-Extra: test
+Requires-Dist: pytest; extra == 'test'
+Description-Content-Type: text/markdown
+
+<img src="./hippoformer-fig6.png" width="400px"></img>
+
+## Hippoformer (wip)
+
+Implementation of [Hippoformer](https://openreview.net/forum?id=hxwV5EubAw), Integrating Hippocampus-inspired Spatial Memory with Transformers
+
+## Citations
+
+```bibtex
+@inproceedings{anonymous2025hippoformer,
+    title   = {Hippoformer: Integrating Hippocampus-inspired Spatial Memory with Transformers},
+    author  = {Anonymous},
+    booktitle = {Submitted to The Fourteenth International Conference on Learning Representations},
+    year    = {2025},
+    url     = {https://openreview.net/forum?id=hxwV5EubAw},
+    note    = {under review}
+}
+```

hippoformer-0.0.12/README.md

@@ -0,0 +1,18 @@
+<img src="./hippoformer-fig6.png" width="400px"></img>
+
+## Hippoformer (wip)
+
+Implementation of [Hippoformer](https://openreview.net/forum?id=hxwV5EubAw), Integrating Hippocampus-inspired Spatial Memory with Transformers
+
+## Citations
+
+```bibtex
+@inproceedings{anonymous2025hippoformer,
+    title   = {Hippoformer: Integrating Hippocampus-inspired Spatial Memory with Transformers},
+    author  = {Anonymous},
+    booktitle = {Submitted to The Fourteenth International Conference on Learning Representations},
+    year    = {2025},
+    url     = {https://openreview.net/forum?id=hxwV5EubAw},
+    note    = {under review}
+}
+```

hippoformer-0.0.12/hippoformer/__init__.py

@@ -0,0 +1,4 @@
+from hippoformer.hippoformer import (
+    PathIntegration,
+    mmTEM
+)

hippoformer-0.0.12/hippoformer/hippoformer.py

@@ -0,0 +1,627 @@
+from __future__ import annotations
+
+import torch
+from torch import nn, Tensor, cat, stack, arange, zeros_like, einsum, tensor
+import torch.nn.functional as F
+from torch.nn import Module
+from torch.jit import ScriptModule, script_method
+from torch.func import vmap, grad, functional_call
+
+from beartype import beartype
+
+from einx import multiply
+from einops import repeat, rearrange, pack, unpack
+from einops.layers.torch import Rearrange
+
+from x_mlps_pytorch import create_mlp
+
+from assoc_scan import AssocScan
+
+# helpers
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+def pack_with_inverse(t, pattern):
+    packed, packed_shape = pack([t], pattern)
+
+    def inverse(out, inv_pattern = None):
+        inv_pattern = default(inv_pattern, pattern)
+        unpacked, = unpack(out, packed_shape, inv_pattern)
+        return unpacked
+
+    return packed, inverse
+
+def l2norm(t):
+    return F.normalize(t, dim = -1)
+
+# Muon - Jordan et al from oss community - applied to the latest version of titans
+
+def newtonschulz5(
+    t,
+    steps = 5,
+    eps = 1e-7,
+    coefs = (3.4445, -4.7750, 2.0315)
+):
+    not_weights = t.ndim <= 3
+
+    if not_weights:
+        return t
+
+    shape = t.shape
+    should_transpose = shape[-2] > shape[-1]
+
+    if should_transpose:
+        t = t.transpose(-1, -2)
+
+    t, inv_pack = pack_with_inverse(t, '* i j')
+    t = t / t.norm(dim = (-1, -2), keepdim = True).clamp(min = eps)
+
+    a, b, c = coefs
+
+    for _ in range(steps):
+        A = t @ t.transpose(-1, -2)
+        B = b * A + c * A @ A
+        t = a * t + B @ t
+
+    if should_transpose:
+        t = t.transpose(-1, -2)
+
+    return inv_pack(t)
+
+# sensory encoder decoder for 2d
+
+grid_sensory_enc_dec = (
+    create_mlp(
+        dim = 32 * 2,
+        dim_in = 9,
+        dim_out = 32,
+        depth = 3,
+    ),
+    create_mlp(
+        dim = 32 * 2,
+        dim_in = 32,
+        dim_out = 9,
+        depth = 3,
+    ),
+)
+
+# sensory encoder decoder for 3d maze
+
+class EncoderPackTime(Module):
+    def __init__(self, fn: Module):
+        super().__init__()
+        self.fn = fn
+
+    def forward(self, x):
+        x = rearrange(x, 'b c t h w -> b t c h w')
+        x, packed_shape = pack([x], '* c h w')
+
+        x = self.fn(x)
+
+        x, = unpack(x, packed_shape, '* d')
+        print(x.shape)
+        return x
+
+class DecoderPackTime(Module):
+    def __init__(self, fn: Module):
+        super().__init__()
+        self.fn = fn
+
+    def forward(self, x):
+        x, packed_shape = pack(x, '* d')
+
+        x = self.fn(x)
+
+        x = unpack(x, packed_shape, '* c h w')
+        x = rearrange(x, 'b t c h w -> b c t h w')
+        return x
+
+maze_sensory_enc_dec = (
+    EncoderPackTime(nn.Sequential(
+        nn.Conv2d(3, 16, 7, 2, padding = 3),
+        nn.ReLU(),
+        nn.Conv2d(16, 32, 3, 2, 1),
+        nn.ReLU(),
+        nn.Conv2d(32, 64, 3, 2, 1),
+        nn.ReLU(),
+        nn.Conv2d(64, 128, 3, 2, 1),
+        nn.ReLU(),
+        Rearrange('b ... -> b (...)'),
+        nn.Linear(2048, 32)
+    )),
+    DecoderPackTime(nn.Sequential(
+        nn.Linear(32, 2048),
+        Rearrange('b (c h w) -> b c h w', c = 128, h = 4),
+        nn.ConvTranspose2d(128, 64, 3, 2, 1, output_padding = (1, 1)),
+        nn.ReLU(),
+        nn.ConvTranspose2d(64, 32, 3, 2, 1, output_padding = (1, 1)),
+        nn.ReLU(),
+        nn.ConvTranspose2d(32, 16, 3, 2, 1, output_padding = (1, 1)),
+        nn.ReLU(),
+        nn.ConvTranspose2d(16, 3, 3, 2, 1, output_padding = (1, 1))
+    ))
+)
+
+# path integration
+
+class RNN(ScriptModule):
+    def __init__(
+        self,
+        dim,
+    ):
+        super().__init__()
+        self.init_hidden = nn.Parameter(torch.randn(1, dim) * 1e-2)
+
+    @script_method
+    def forward(
+        self,
+        transitions: Tensor,
+        hidden: Tensor | None = None
+    ) -> Tensor:
+
+        batch, seq_len = transitions.shape[:2]
+
+        if hidden is None:
+            hidden = l2norm(self.init_hidden)
+            hidden = hidden.expand(batch, -1)
+
+        hiddens: list[Tensor] = []
+
+        for i in range(seq_len):
+            transition = transitions[:, i]
+
+            hidden = einsum('b i, b i j -> b j', hidden, transition)
+            hidden = F.relu(hidden)
+            hidden = l2norm(hidden)
+
+            hiddens.append(hidden)
+
+        return stack(hiddens, dim = 1)
+
+class PathIntegration(Module):
+    def __init__(
+        self,
+        dim_action,
+        dim_structure,
+        mlp_hidden_dim = None,
+        mlp_depth = 2
+    ):
+        # they use the same approach from Ruiqi Gao's paper from 2021
+        super().__init__()
+
+        self.init_structure = nn.Parameter(torch.randn(dim_structure))
+
+        self.to_transitions = create_mlp(
+            default(mlp_hidden_dim,  dim_action * 4),
+            dim_in = dim_action,
+            dim_out = dim_structure * dim_structure,
+            depth = mlp_depth
+        )
+
+        self.mlp_out_to_weights = Rearrange('... (i j) -> ... i j', j = dim_structure)
+
+        self.rnn = RNN(dim_structure)
+
+    def forward(
+        self,
+        actions,                 # (b n d)
+        prev_structural = None   # (b n d) | (b d)
+    ):
+        batch = actions.shape[0]
+
+        transitions = self.to_transitions(actions)
+        transitions = self.mlp_out_to_weights(transitions)
+
+        if exists(prev_structural) and prev_structural.ndim == 3:
+            prev_structural = prev_structural[:, -1]
+
+        return self.rnn(transitions, prev_structural)
+
+# custom transformer proposed by James Whittington that bridges to hippocampal models with a few twists
+
+# the mmTEM can be seen as a linear attention / TTT variant of what he proposed
+# needed for the baseline as well as the parallel block to bolster local time prediction
+
+# https://arxiv.org/abs/2112.04035
+
+def FeedForward(dim, mult = 4.):
+    dim_inner = int(dim * mult)
+    return nn.Sequential(
+        nn.Linear(dim, dim_inner),
+        nn.GELU(),
+        nn.Linear(dim_inner, dim)
+    )
+
+class Attention(Module):
+    def __init__(
+        self,
+        dim_q,
+        dim_kv,
+        window_size,
+        dim_head = 64,
+        heads = 8,
+    ):
+        super().__init__()
+        dim_inner = dim_head * heads
+        self.scale = dim_head ** -0.5
+
+        self.to_queries = nn.Linear(dim_q, dim_inner, bias = False)
+        self.to_key_values = nn.Linear(dim_kv, dim_inner * 2, bias = False)
+
+        self.split_heads = Rearrange('b n (h d) -> b h n d', h = heads)
+        self.merge_heads = Rearrange('b h n d -> b n (h d)')
+
+        self.window_size = window_size
+
+        self.to_out = nn.Linear(dim_inner, dim_q, bias = False)
+        self.attn_head_sink = nn.Parameter(torch.randn(heads) * 1e-2) # needed as the diagonal is masked out, and for attention sink
+
+    def forward(
+        self,
+        queries_input,
+        key_values_input,
+        kv_cache = None
+    ):
+        batch, seq_len, device = *queries_input.shape[:2], queries_input.device
+
+        q = self.to_queries(queries_input)
+
+        k, v = self.to_key_values(key_values_input).chunk(2, dim = -1)
+
+        q, k, v = tuple(self.split_heads(t) for t in (q, k, v))
+
+        if exists(kv_cache):
+            ck, cv = kv_cache
+            k = cat((ck, k), dim = -2)
+            v = cat((cv, v), dim = -2)
+
+        q = q * self.scale
+
+        sim = einsum('b h i d, b h j d -> b h i j', q, k)
+
+        # the diagonal is masked out
+
+        i, j = sim.shape[-2:]
+
+        j_seq = arange(j, device = device)[:, None]
+        i_seq = arange(i, device = device)[None, :] + (j - i)
+
+        windowed_causal_mask_without_diagonal = (i_seq > j_seq) & ((i_seq - j_seq) <= self.window_size)
+
+        sim = sim.masked_fill(windowed_causal_mask_without_diagonal, -torch.finfo(sim.dtype).max)
+
+        # attention sink, for token as well as for attention sinking - from gpt-oss
+
+        attn_sink = repeat(self.attn_head_sink, 'h -> b h i 1', b = batch, i = seq_len)
+
+        sim = cat((attn_sink, sim), dim = -1)
+
+        attn = sim.softmax(dim = -1)
+
+        attn = attn[..., 1:] # remove sink
+
+        # aggregate
+
+        out = einsum('b h i j, b h j d -> b h i d', attn, v)
+
+        out = self.merge_heads(out)
+
+        return self.to_out(out), stack((k, v))
+
+class TEMTransformerBlock(Module):
+    def __init__(
+        self,
+        dim_structure,
+        dim_encoded_sensory,
+        dim_head = 64,
+        heads = 8,
+        ff_expansion_factor = 4.,
+        window_size = 64
+    ):
+        super().__init__()
+
+        self.attn = Attention(dim_structure, dim_structure + dim_encoded_sensory, window_size, dim_head = dim_head, heads = heads)
+        self.ff = FeedForward(dim_structure, ff_expansion_factor)
+
+        self.window_size = window_size
+
+    def forward(
+        self,
+        structural_codes,
+        encoded_sensory,
+        kv_cache = None
+    ):
+        structure_and_sensory = cat((structural_codes, encoded_sensory), dim = -1)
+
+        retrieved, next_kv_cache = self.attn(structural_codes, structure_and_sensory, kv_cache = kv_cache)
+
+        x = retrieved + structural_codes
+
+        x = self.ff(x) + x
+
+        next_kv_cache = next_kv_cache[:, -self.window_size:]
+
+        return x, next_kv_cache
+
+# proposed mmTEM
+
+class mmTEM(Module):
+    @beartype
+    def __init__(
+        self,
+        dim,
+        *,
+        sensory_encoder_decoder: tuple[Module, Module],
+        dim_sensory,
+        dim_action,
+        dim_encoded_sensory,
+        dim_structure,
+        meta_mlp_depth = 2,
+        decoder_mlp_depth = 2,
+        structure_variance_pred_mlp_depth = 2,
+        path_integrate_kwargs: dict = dict(),
+        loss_weight_generative = 1.,
+        loss_weight_inference = 1.,
+        loss_weight_consistency = 1.,
+        loss_weight_relational = 1.,
+        integration_ratio_learned = True,
+        muon_update = False,
+        assoc_scan_kwargs: dict = dict()
+    ):
+        super().__init__()
+
+        # sensory
+
+        sensory_encoder, sensory_decoder = sensory_encoder_decoder
+
+        self.sensory_encoder = sensory_encoder
+        self.sensory_decoder = sensory_decoder
+
+        dim_joint_rep = dim_encoded_sensory + dim_structure
+
+        self.dim_encoded_sensory = dim_encoded_sensory
+        self.dim_structure = dim_structure
+        self.joint_dims = (dim_structure, dim_encoded_sensory)
+
+        # path integrator
+
+        self.path_integrator = PathIntegration(
+            dim_action = dim_action,
+            dim_structure = dim_structure,
+            **path_integrate_kwargs
+        )
+
+        # meta mlp related
+
+        self.to_queries = nn.Linear(dim_joint_rep, dim, bias = False)
+        self.to_keys = nn.Linear(dim_joint_rep, dim, bias = False)
+        self.to_values = nn.Linear(dim_joint_rep, dim, bias = False)
+
+        self.to_learned_optim_hparams = nn.Linear(dim_joint_rep, 3, bias = False) # for learning rate, forget gate, and momentum
+        self.assoc_scan = AssocScan(*assoc_scan_kwargs)
+
+        self.meta_memory_mlp = create_mlp(
+            dim = dim * 2,
+            depth = meta_mlp_depth,
+            dim_in = dim,
+            dim_out = dim,
+            activation = nn.ReLU()
+        )
+
+        def forward_with_mse_loss(params, keys, values):
+            pred = functional_call(self.meta_memory_mlp, params, keys)
+            return F.mse_loss(pred, values)
+
+        grad_fn = grad(forward_with_mse_loss)
+
+        self.per_sample_grad_fn = vmap(vmap(grad_fn, in_dims = (None, 0, 0)), in_dims = (0, 0, 0))
+
+        # mlp decoder (from meta mlp output to joint)
+
+        self.memory_output_decoder = create_mlp(
+            dim = dim * 2,
+            dim_in = dim,
+            dim_out = dim_joint_rep,
+            depth = decoder_mlp_depth,
+            activation = nn.ReLU()
+        )
+
+        # the mlp that predicts the variance for the structural code
+        # for correcting the generated structural code modeling the feedback from HC to MEC
+
+        self.structure_variance_pred_mlp = create_mlp(
+            dim = dim_structure * 2,
+            dim_in = dim_structure * 2 + 1,
+            dim_out = dim_structure,
+            depth = structure_variance_pred_mlp_depth
+        )
+
+        # loss related
+
+        self.loss_weight_generative = loss_weight_generative
+        self.loss_weight_inference = loss_weight_inference
+        self.loss_weight_relational = loss_weight_relational
+        self.loss_weight_consistency = loss_weight_consistency
+        self.register_buffer('zero', tensor(0.), persistent = False)
+
+        # update with muon
+
+        self.muon_update = muon_update
+
+        # there is an integration ratio for error correction, but unclear what value this is fixed to or whether it is learned
+
+        self.integration_ratio = nn.Parameter(tensor(0.), requires_grad = integration_ratio_learned)
+
+    def init_params_and_momentum(
+        self,
+        batch_size
+    ):
+
+        params_dict = dict(self.meta_memory_mlp.named_parameters())
+
+        params = {name: repeat(param, '... -> b ...', b = batch_size) for name, param in params_dict.items()}
+
+        momentums = {name: zeros_like(param) for name, param in params.items()}
+
+        return params, momentums
+
+    def retrieve(
+        self,
+        structural_codes,
+        encoded_sensory
+    ):
+        joint = cat((structural_codes, encoded_sensory), dim = -1)
+
+        queries = self.to_queries(joint)
+
+        retrieved = self.meta_memory_mlp(queries)
+
+        return self.memory_output_decoder(retrieved).split(self.joint_dims, dim = -1)
+
+    def forward(
+        self,
+        sensory,
+        actions,
+        memory_mlp_params = None,
+        return_losses = False,
+        return_memory_mlp_params = False
+    ):
+        batch = actions.shape[0]
+
+        structural_codes = self.path_integrator(actions)
+
+        encoded_sensory = self.sensory_encoder(sensory)
+
+        # 1. first have the structure code be able to fetch from the meta memory mlp
+
+        decoded_gen_structure, decoded_encoded_sensory = self.retrieve(structural_codes, zeros_like(encoded_sensory))
+
+        decoded_sensory = self.sensory_decoder(decoded_encoded_sensory)
+
+        generative_pred_loss = F.mse_loss(sensory, decoded_sensory)
+
+        # 2. relational
+
+        # 2a. structure from content
+
+        decoded_structure, decoded_encoded_sensory = self.retrieve(zeros_like(structural_codes), encoded_sensory)
+
+        structure_from_content_loss = F.mse_loss(decoded_structure, structural_codes)
+
+        # 2b. structure from structure
+
+        decoded_structure, decoded_encoded_sensory = self.retrieve(zeros_like(structural_codes), encoded_sensory)
+
+        structure_from_structure_loss = F.mse_loss(decoded_structure, structural_codes)
+
+        relational_loss = structure_from_content_loss + structure_from_structure_loss
+
+        # 3. consistency - modeling a feedback system from hippocampus to path integration
+
+        corrected_structural_code, corrected_encoded_sensory = self.retrieve(decoded_gen_structure, encoded_sensory)
+
+        sensory_sse = (corrected_encoded_sensory - encoded_sensory).norm(dim = -1, keepdim = True).pow(2)
+
+        pred_variance = self.structure_variance_pred_mlp(cat((corrected_structural_code, decoded_gen_structure, sensory_sse), dim = -1))
+
+        inf_structural_code = decoded_gen_structure + (corrected_structural_code - decoded_gen_structure) * self.integration_ratio.sigmoid() * pred_variance
+
+        consistency_loss = F.mse_loss(decoded_gen_structure, inf_structural_code)
+
+        # 4. final inference loss
+
+        final_structural_code, inf_encoded_sensory = self.retrieve(inf_structural_code, zeros_like(encoded_sensory))
+
+        decoded_inf_sensory = self.sensory_decoder(inf_encoded_sensory)
+
+        inference_pred_loss = F.mse_loss(sensory, decoded_inf_sensory)
+
+        # 5. store the final structural code from step 4 + encoded sensory
+
+        joint_code_to_store = cat((final_structural_code, encoded_sensory), dim = -1)
+
+        keys = self.to_keys(joint_code_to_store)
+        values = self.to_values(joint_code_to_store)
+
+        lr, forget, beta = self.to_learned_optim_hparams(joint_code_to_store).unbind(dim = -1)
+
+        if exists(memory_mlp_params):
+            params, momentums = memory_mlp_params
+        else:
+            params, momentums = self.init_params_and_momentum(batch)
+
+        # store by getting gradients of mse loss of keys and values
+
+        grads = self.per_sample_grad_fn(params, keys, values)
+
+        # update the meta mlp parameters and momentums
+
+        next_params = dict()
+        next_momentum = dict()
+
+        for (
+            (key, param),
+            (_, grad),
+            (_, momentum)
+        ) in zip(
+            params.items(),
+            grads.items(),
+            momentums.items()
+        ):
+
+            grad, inverse_pack = pack_with_inverse(grad, 'b t *')
+
+            grad = multiply('b t ..., b t', grad, lr)
+
+            expanded_beta = repeat(beta, 'b t -> b t w', w = grad.shape[-1])
+
+            update = self.assoc_scan(grad, expanded_beta.sigmoid(), momentum)
+
+            # store next momentum
+
+            next_momentum[key] = update[:, -1]
+
+            # maybe muon
+
+            if self.muon_update:
+                update = newtonschulz5(update)
+
+            # with forget gating
+
+            expanded_forget = repeat(forget, 'b t -> b t w', w = grad.shape[-1])
+
+            acc_update = self.assoc_scan(-update, expanded_forget.sigmoid(), param)
+
+            acc_update = inverse_pack(acc_update)
+
+            # set the next params and momentum, which can be passed back in
+
+            next_params[key] =  acc_update[:, -1]
+
+        # losses
+
+        total_loss = (
+            generative_pred_loss * self.loss_weight_generative +
+            relational_loss * self.loss_weight_relational +
+            consistency_loss * self.loss_weight_consistency +
+            inference_pred_loss * self.loss_weight_inference
+        )
+
+        losses = (
+            generative_pred_loss,
+            relational_loss,
+            consistency_loss,
+            inference_pred_loss
+        )
+
+        if return_memory_mlp_params:
+            return next_params, next_momentum
+
+        if not return_losses:
+            return total_loss
+
+        return total_loss, losses

hippoformer-0.0.12/pyproject.toml

@@ -0,0 +1,62 @@
+[project]
+name = "hippoformer"
+version = "0.0.12"
+description = "hippoformer"
+authors = [
+    { name = "Phil Wang", email = "lucidrains@gmail.com" }
+]
+readme = "README.md"
+requires-python = ">= 3.9"
+license = { file = "LICENSE" }
+keywords = [
+    'artificial intelligence',
+    'deep learning',
+    'memory',
+    'hippocampus'
+]
+
+classifiers=[
+    'Development Status :: 4 - Beta',
+    'Intended Audience :: Developers',
+    'Topic :: Scientific/Engineering :: Artificial Intelligence',
+    'License :: OSI Approved :: MIT License',
+    'Programming Language :: Python :: 3.9',
+]
+
+dependencies = [
+    "assoc-scan",
+    "beartype",
+    "einx>=0.3.0",
+    "einops>=0.8.1",
+    "torch>=2.4",
+    "x-mlps-pytorch",
+]
+
+[project.urls]
+Homepage = "https://pypi.org/project/hippoformer/"
+Repository = "https://github.com/lucidrains/hippoformer"
+
+[project.optional-dependencies]
+examples = []
+test = [
+    "pytest"
+]
+
+[tool.pytest.ini_options]
+pythonpath = [
+  "."
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.rye]
+managed = true
+dev-dependencies = []
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.build.targets.wheel]
+packages = ["hippoformer"]

hippoformer-0.0.12/tests/test_hippoformer.py

@@ -0,0 +1,79 @@
+import pytest
+param = pytest.mark.parametrize
+
+import torch
+
+def test_path_integrate():
+    from hippoformer.hippoformer import PathIntegration
+
+    path_integrator = PathIntegration(32, 64)
+
+    actions = torch.randn(2, 16, 32)
+
+    structure_codes = path_integrator(actions)
+    structure_codes = path_integrator(actions, structure_codes) # pass in previous structure codes, it will auto use the last one as hidden and pass it to the RNN
+
+    assert structure_codes.shape == (2, 16, 64)
+
+@param('sensory_type', ('naive', '2d', '3d'))
+@param('muon_update', (True, False))
+def test_mm_tem(
+    sensory_type,
+    muon_update
+):
+    import torch
+    from hippoformer.hippoformer import mmTEM
+
+    from torch.nn import Linear
+
+    if sensory_type == 'naive':
+        enc_dec = (
+            Linear(11, 32),
+            Linear(32, 11)
+        )
+        sensory = torch.randn(2, 16, 11)
+
+    elif sensory_type == '2d':
+
+        from hippoformer.hippoformer import grid_sensory_enc_dec
+
+        enc_dec = grid_sensory_enc_dec
+        sensory = torch.randn(2, 16, 9)
+
+    elif sensory_type == '3d':
+
+        from hippoformer.hippoformer import maze_sensory_enc_dec
+
+        enc_dec = maze_sensory_enc_dec
+
+        sensory = torch.randn(2, 3, 16, 64, 64)
+
+    model = mmTEM(
+        dim = 32,
+        sensory_encoder_decoder = enc_dec,
+        dim_sensory = 11,
+        dim_action = 7,
+        dim_structure = 32,
+        dim_encoded_sensory = 32,
+        muon_update = muon_update
+    )
+
+    actions = torch.randn(2, 16, 7)
+
+    next_params = model(sensory, actions, return_memory_mlp_params = True)
+    next_params = model(sensory, actions, memory_mlp_params = next_params, return_memory_mlp_params = True)
+
+    loss = model(sensory, actions, memory_mlp_params = next_params)
+    loss.backward()
+
+def test_tem_t():
+    from hippoformer.hippoformer import TEMTransformerBlock
+
+    block = TEMTransformerBlock(32, 16, window_size = 3)
+
+    structural_codes = torch.randn(1, 7, 32)
+    encoded_sensory = torch.randn(1, 7, 16)
+
+    pred, kv_cache = block(structural_codes, encoded_sensory)
+
+    assert pred.shape == (1, 7, 32)