self-play-pretrain-zero-data 0.0.1__tar.gz

self_play_pretrain_zero_data-0.0.1/.gitignore

@@ -0,0 +1,220 @@
+*.sh
+
+# 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
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# 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/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml

self_play_pretrain_zero_data-0.0.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 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.

self_play_pretrain_zero_data-0.0.1/PKG-INFO

@@ -0,0 +1,77 @@
+Metadata-Version: 2.5
+Name: self-play-pretrain-zero-data
+Version: 0.0.1
+Summary: Implementation of Self-Play Pretraining with Zero Data
+Project-URL: Homepage, https://pypi.org/project/self-play-pretrain-zero-data/
+Project-URL: Repository, https://github.com/lucidrains/self-play-pretrain-zero-data
+Author-email: Phil Wang <lucidrains@gmail.com>
+License: MIT License
+        
+        Copyright (c) 2026 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,pretraining,self-play,synthetic data
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Requires-Dist: einops>=0.8.1
+Requires-Dist: einx>=0.3.0
+Requires-Dist: torch-einops-utils>=0.0.29
+Requires-Dist: torch>=2.5
+Requires-Dist: x-transformers
+Provides-Extra: examples
+Provides-Extra: test
+Requires-Dist: pytest; extra == 'test'
+Description-Content-Type: text/markdown
+
+## self-play-pretrain-zero-data (wip)
+
+Implementation of [Self-Play Pretraining with Zero Data](https://arxiv.org/abs/2609.30063)
+
+[Paper Review from @hu-po](https://www.youtube.com/watch?v=mGMiiPpWBSo)
+
+## Citations
+
+```bibtex
+@misc{cowsik2026selfplaypretrainingzerodata,
+    title    = {Self-Play Pretraining with Zero Data},
+    author   = {Aditya Cowsik and Kfir Dolev and Michael Y. Li and G. Bruno De Luca and Nourya Cohen and Noah D. Goodman and Yoav Levine},
+    year     = {2026},
+    eprint   = {2609.30063},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.AI},
+    url      = {https://arxiv.org/abs/2609.30063},
+}
+```
+
+```bibtex
+@misc{bloem2025universalpretrainingiteratedrandom,
+    title   = {Universal pre-training by iterated random computation},
+    author  = {Peter Bloem},
+    year    = {2025},
+    eprint  = {2506.20057},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.LG},
+    url     = {https://arxiv.org/abs/2506.20057},
+}
+```

self_play_pretrain_zero_data-0.0.1/README.md

@@ -0,0 +1,31 @@
+## self-play-pretrain-zero-data (wip)
+
+Implementation of [Self-Play Pretraining with Zero Data](https://arxiv.org/abs/2609.30063)
+
+[Paper Review from @hu-po](https://www.youtube.com/watch?v=mGMiiPpWBSo)
+
+## Citations
+
+```bibtex
+@misc{cowsik2026selfplaypretrainingzerodata,
+    title    = {Self-Play Pretraining with Zero Data},
+    author   = {Aditya Cowsik and Kfir Dolev and Michael Y. Li and G. Bruno De Luca and Nourya Cohen and Noah D. Goodman and Yoav Levine},
+    year     = {2026},
+    eprint   = {2609.30063},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.AI},
+    url      = {https://arxiv.org/abs/2609.30063},
+}
+```
+
+```bibtex
+@misc{bloem2025universalpretrainingiteratedrandom,
+    title   = {Universal pre-training by iterated random computation},
+    author  = {Peter Bloem},
+    year    = {2025},
+    eprint  = {2506.20057},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.LG},
+    url     = {https://arxiv.org/abs/2506.20057},
+}
+```

self_play_pretrain_zero_data-0.0.1/pyproject.toml

@@ -0,0 +1,63 @@
+[project]
+name = "self-play-pretrain-zero-data"
+version = "0.0.1"
+description = "Implementation of Self-Play Pretraining with Zero Data"
+authors = [
+    { name = "Phil Wang", email = "lucidrains@gmail.com" }
+]
+readme = "README.md"
+requires-python = ">= 3.10"
+license = { file = "LICENSE" }
+keywords = [
+    'artificial intelligence',
+    'deep learning',
+    'self-play',
+    'pretraining',
+    'synthetic data'
+]
+
+classifiers=[
+    'Development Status :: 4 - Beta',
+    'Intended Audience :: Developers',
+    'Topic :: Scientific/Engineering :: Artificial Intelligence',
+    'License :: OSI Approved :: MIT License',
+    'Programming Language :: Python :: 3.10',
+]
+
+dependencies = [
+    "einx>=0.3.0",
+    "einops>=0.8.1",
+    "torch>=2.5",
+    "torch-einops-utils>=0.0.29",
+    "x-transformers"
+]
+
+[project.urls]
+Homepage = "https://pypi.org/project/self-play-pretrain-zero-data/"
+Repository = "https://github.com/lucidrains/self-play-pretrain-zero-data"
+
+[project.optional-dependencies]
+examples = []
+test = [
+    "pytest"
+]
+
+[tool.pytest.ini_options]
+pythonpath = [
+    "."
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build]
+include = [
+    "self_play_pretrain_zero_data",
+    "pyproject.toml",
+    "README.md",
+    "LICENSE"
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["self_play_pretrain_zero_data"]

self_play_pretrain_zero_data-0.0.1/self_play_pretrain_zero_data/__init__.py

@@ -0,0 +1,6 @@
+from self_play_pretrain_zero_data.self_play import (
+    SelfPlay,
+    ProgramExecutor
+)
+
+from self_play_pretrain_zero_data.brainfuck import Brainfuck

self_play_pretrain_zero_data-0.0.1/self_play_pretrain_zero_data/brainfuck.py

@@ -0,0 +1,96 @@
+from __future__ import annotations
+
+import random
+from itertools import chain, repeat
+
+from self_play_pretrain_zero_data.self_play import ProgramExecutor
+
+# single byte macro instructions from appendix E (table 4)
+
+MACROS = {
+    'Z': '[-]',
+    'R': '[->+<]',
+    'L': '[->+++<]',
+    'N': '[-<->]',
+    'C': '[->+>+<<]',
+    'G': '[>]',
+    'H': '[<]',
+    'W': '[[-]>+<]',
+    'V': '[.>]',
+    'X': '[-]++++++++++++++++',
+}
+
+
+def bracket_jumps(code):
+    jumps, stack = {}, []
+
+    for i, c in enumerate(code):
+        if c == '[':
+            stack.append(i)
+        elif c == ']' and stack:
+            left = stack.pop()
+            jumps[left] = i
+            jumps[i] = left
+
+    return jumps
+
+class Brainfuck(ProgramExecutor):
+    """appendix E execution semantics - every string is executable
+
+    - circular tape, cells modulo `modulus`
+    - unmatched brackets are no-ops
+    - `,` reads the input string, then uniform random bytes (random input tape omega)
+    - halts at `F`, the step budget, or the output cap
+    """
+
+    def __init__(
+        self,
+        tape_size = 1024,
+        modulus = 256,
+        max_steps = 100_000,
+        max_output_len = 1024
+    ):
+        self.tape_size = tape_size
+        self.modulus = modulus
+        self.max_steps = max_steps
+        self.max_output_len = max_output_len
+
+    def __call__(
+        self,
+        program: str,
+        input: str | bytes = '',
+        seed: int | None = None
+    ) -> str:
+        rng = random.Random(seed)
+
+        code = ''.join(MACROS.get(c, c) for c in program.partition('F')[0])
+        jumps = bracket_jumps(code)
+
+        input_bytes = input.encode('latin-1') if isinstance(input, str) else input
+
+        reads = chain(
+            (b % self.modulus for b in input_bytes),
+            (rng.randrange(self.modulus) for _ in repeat(None))
+        )
+
+        tape = [0] * self.tape_size
+        out = bytearray()
+        pc = ptr = 0
+
+        for _ in range(self.max_steps):
+            if pc >= len(code) or len(out) >= self.max_output_len:
+                break
+
+            match code[pc]:
+                case '>': ptr = (ptr + 1) % self.tape_size
+                case '<': ptr = (ptr - 1) % self.tape_size
+                case '+': tape[ptr] = (tape[ptr] + 1) % self.modulus
+                case '-': tape[ptr] = (tape[ptr] - 1) % self.modulus
+                case '.': out.append(tape[ptr])
+                case ',': tape[ptr] = next(reads)
+                case '[' if not tape[ptr]: pc = jumps.get(pc, pc)
+                case ']' if tape[ptr]: pc = jumps.get(pc, pc)
+
+            pc += 1
+
+        return out.decode('latin-1')

self_play_pretrain_zero_data-0.0.1/self_play_pretrain_zero_data/self_play.py

@@ -0,0 +1,42 @@
+from __future__ import annotations
+from typing import Iterable
+from abc import ABC, abstractmethod
+
+from torch.nn import Module
+
+# executor
+
+class ProgramExecutor(ABC):
+    """maps a program string to its output string"""
+
+    @abstractmethod
+    def __call__(self, program: str, **kwargs) -> str:
+        raise NotImplementedError
+
+# main class
+
+class SelfPlay(Module):
+    def __init__(
+        self,
+        generator: Module,
+        learner: Module,
+        executor: ProgramExecutor
+    ):
+        super().__init__()
+        self.generator = generator
+        self.learner = learner
+        self.executor = executor
+
+    def execute(
+        self,
+        programs: str | Iterable[str],
+        **kwargs
+    ) -> str | list[str]:
+
+        if isinstance(programs, str):
+            return self.executor(programs, **kwargs)
+
+        return [self.executor(program, **kwargs) for program in programs]
+
+    def forward(self, *args, **kwargs):
+        raise NotImplementedError