axiomos


Nameaxiomos JSON
Version 0.1.18 PyPI version JSON
download
home_pageNone
SummaryA lightweight AXIR playground to run numerical kernels across CPU and OpenCL
upload_time2025-10-13 23:00:47
maintainerNone
docs_urlNone
authorAxiomos
requires_python>=3.10
licenseMIT
keywords opencl kernels gemm axir numpy runtime ai
VCS
bugtrack_url
requirements No requirements were recorded.
Travis-CI No Travis.
coveralls test coverage No coveralls. 
            # AXIOMOS — Public Showcase (Minimal)
[PyPI](https://pypi.org/project/axiomos/) • [GitHub](https://github.com/Aidenkuro10/axiomos-vitrine)  
![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)

> 🔒 Public showcase build — proves the critical path: **AXIR JSON → CPU/OpenCL execution → numeric verification**.  
> The private core (optimizers, scheduler, advanced kernels, **cryptographic trust**) remains **under NDA**.

---

## 🚀 TL;DR — 3-command demo

### Windows (PowerShell)
```powershell
pip install axiomos; or pip install "axiomos>=0.1.18"
axiomos-doctor
axiomos-demo

# optional
axiomos-devices
axiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0
macOS / Linux (zsh/bash)

pip install "axiomos>=0.1.18"
axiomos-doctor
axiomos-demo

# optional
axiomos-devices
axiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0

axiomos-doctor prints versions, attempts OpenCL discovery, and runs a softmax 8×8 sanity test.
axiomos-demo creates a tiny AXIR JSON and verifies CPU ↔ OPENCL.
If OpenCL isn’t available, it falls back to OPENCL(cpu-fallback) and still verifies ALLCLOSE (strict parity).


🆘 If a command is “not recognized” (universal fallback)
Use the module form (bypasses PATH issues and mixed Python installs):

Windows (PowerShell)
python -m pip install -U "axiomos>=0.1.18"
python -m axiomos.cli.axir_doctor
python -m axiomos.cli.quick_demo
python -m axiomos.cli.devices
python -m axiomos.smoke --size 512 --warmup 3 --repeat 30 --seed 0

macOS / Linux (zsh/bash)
python3 -m pip install -U "axiomos>=0.1.18"
python3 -m axiomos.cli.axir_doctor
python3 -m axiomos.cli.quick_demo
python3 -m axiomos.cli.devices
python3 -m axiomos.smoke --size 512 --warmup 3 --repeat 30 --seed 0

On Windows, if you use a virtualenv, activate it first:
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install "axiomos>=0.1.18"
axiomos-doctor
axiomos-demo


🎯 What this proves (public)
Portability — the same .axir.json runs on CPU and OpenCL (or a graceful CPU fallback).

Verifiability — strict numeric parity (ALLCLOSE) with shapes, max_abs_err, timings.

Reproducibility — seedable runs, explicit versions printed.

UX — install → run → verify in minutes; no private code exposed.

Not included here (private): optimizer passes, scheduler, advanced kernels, full operator coverage, vendor-tuned implementations, and the full trust pipeline (content-addressed artifacts, Ed25519 signatures, attestations).
➡️ A full private demo is available under NDA.

🔧 Useful commands
List OpenCL devices (if PyOpenCL + drivers are present)

axiomos-devices
# universal fallback:
python -m axiomos.cli.devices
Verify a fixture (example: Softmax 8×8)
python -m axiomos.verify examples/softmax2d_small.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0
Output includes: SHAPES, max_abs_err, ALLCLOSE (atol=1e-6, rtol=1e-5), CPU & OPENCL timings.
Note: the public OpenCL path is minimal (not optimized) — correctness, not performance.

Tiny smoke (latency indicator)
axiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0
# fallback:
python -m axiomos.smoke --size 512 --warmup 3 --repeat 30 --seed 0

🧪 Optional: PyTorch → AXIR mini-export + verify
Install PyTorch (it’s large; keep optional for demos):
pip install torch
Export a tiny model to AXIR JSON:
axiomos-export-torch
# fallback:

This creates:
examples/pytorch_softmax.axir.json
Verify CPU ↔ OPENCL on the exported AXIR:

python -m axiomos.verify examples/pytorch_softmax.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0
As with other demos: if OpenCL isn’t available, verification falls back to OPENCL(cpu-fallback) and still checks numeric parity.

🧰 If you cloned the repo (extra fixtures)
Generate fixtures:

python make_fixtures.py
Creates:
examples/
 ├─ vector_add_small.axir.json
 └─ softmax2d_small.axir.json
Verify:

# CPU ↔ CPU (Hello AXIR)
axiomos-verify examples/vector_add_small.axir.json --buffer hC --backend-a cpu --backend-b cpu --seed 0

# CPU ↔ OpenCL (Softmax 8×8)
python -m axiomos.verify examples/softmax2d_small.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0

💡 Common errors & fixes
“Command not found” (e.g., axiomos-devices):
Use module form (works everywhere):
python -m axiomos.cli.devices (Windows) or python3 -m axiomos.cli.devices (macOS/Linux).
On Windows, if using a venv, activate it before running commands.

“No matching distribution found for axiomos”
Python is likely too old. Use Python 3.10+. On macOS:
brew install python@3.11
/opt/homebrew/bin/python3.11 -m venv .venv   # Apple Silicon
# or: /usr/local/bin/python3.11 -m venv .venv # Intel
source .venv/bin/activate
python -m pip install -U pip setuptools wheel
python -m pip install -U "axiomos>=0.1.15"

Apple Silicon arch mismatch (Rosetta vs arm64):
usr/bin/arch -arm64 python3 -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
python -m pip install -U "axiomos>=0.1.15"
Corporate proxy / SSL errors:
Configure pip to use your proxy or trusted certs, or install inside a VPN/approved network. (Keep the module form.)

PyTorch install is slow:
That’s expected (large wheels). Keep PyTorch optional for the demo.

Why AXIOMOS
AXIR (Axiomos IR) is a universal, hardware-agnostic IR aimed at:

Portability — compile once, run on CPU, GPU, and accelerators.

Determinism & Reproducibility — numerically verifiable parity across backends.

Trust — measured parity today; cryptographic provenance/signatures in the private build.

Longevity — a stable IR beyond today’s frameworks and vendor APIs.

This public showcase is intentionally minimal: it proves AXIR JSON → multi-backend execution → verification without exposing the private core.

License
MIT (public showcase only). The full runtime remains proprietary.

Raw data

            {
    "_id": null,
    "home_page": null,
    "name": "axiomos",
    "maintainer": null,
    "docs_url": null,
    "requires_python": ">=3.10",
    "maintainer_email": null,
    "keywords": "opencl, kernels, gemm, axir, numpy, runtime, AI",
    "author": "Axiomos",
    "author_email": null,
    "download_url": "https://files.pythonhosted.org/packages/6e/ec/7f2e40973f7809a098ab86fef7e3b4576829d19b336aaf5c905a15d42d4a/axiomos-0.1.18.tar.gz",
    "platform": null,
    "description": "# AXIOMOS \u2014 Public Showcase (Minimal)\r\n[PyPI](https://pypi.org/project/axiomos/) \u2022 [GitHub](https://github.com/Aidenkuro10/axiomos-vitrine)  \r\n![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)\r\n\r\n> \ud83d\udd12 Public showcase build \u2014 proves the critical path: **AXIR JSON \u2192 CPU/OpenCL execution \u2192 numeric verification**.  \r\n> The private core (optimizers, scheduler, advanced kernels, **cryptographic trust**) remains **under NDA**.\r\n\r\n---\r\n\r\n## \ud83d\ude80 TL;DR \u2014 3-command demo\r\n\r\n### Windows (PowerShell)\r\n```powershell\r\npip install axiomos; or pip install \"axiomos>=0.1.18\"\r\naxiomos-doctor\r\naxiomos-demo\r\n\r\n# optional\r\naxiomos-devices\r\naxiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0\r\nmacOS / Linux (zsh/bash)\r\n\r\npip install \"axiomos>=0.1.18\"\r\naxiomos-doctor\r\naxiomos-demo\r\n\r\n# optional\r\naxiomos-devices\r\naxiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0\r\n\r\naxiomos-doctor prints versions, attempts OpenCL discovery, and runs a softmax 8\u00d78 sanity test.\r\naxiomos-demo creates a tiny AXIR JSON and verifies CPU \u2194 OPENCL.\r\nIf OpenCL isn\u2019t available, it falls back to OPENCL(cpu-fallback) and still verifies ALLCLOSE (strict parity).\r\n\r\n\r\n\ud83c\udd98 If a command is \u201cnot recognized\u201d (universal fallback)\r\nUse the module form (bypasses PATH issues and mixed Python installs):\r\n\r\nWindows (PowerShell)\r\npython -m pip install -U \"axiomos>=0.1.18\"\r\npython -m axiomos.cli.axir_doctor\r\npython -m axiomos.cli.quick_demo\r\npython -m axiomos.cli.devices\r\npython -m axiomos.smoke --size 512 --warmup 3 --repeat 30 --seed 0\r\n\r\nmacOS / Linux (zsh/bash)\r\npython3 -m pip install -U \"axiomos>=0.1.18\"\r\npython3 -m axiomos.cli.axir_doctor\r\npython3 -m axiomos.cli.quick_demo\r\npython3 -m axiomos.cli.devices\r\npython3 -m axiomos.smoke --size 512 --warmup 3 --repeat 30 --seed 0\r\n\r\nOn Windows, if you use a virtualenv, activate it first:\r\npython -m venv .venv\r\n.\\.venv\\Scripts\\Activate.ps1\r\npip install \"axiomos>=0.1.18\"\r\naxiomos-doctor\r\naxiomos-demo\r\n\r\n\r\n\ud83c\udfaf What this proves (public)\r\nPortability \u2014 the same .axir.json runs on CPU and OpenCL (or a graceful CPU fallback).\r\n\r\nVerifiability \u2014 strict numeric parity (ALLCLOSE) with shapes, max_abs_err, timings.\r\n\r\nReproducibility \u2014 seedable runs, explicit versions printed.\r\n\r\nUX \u2014 install \u2192 run \u2192 verify in minutes; no private code exposed.\r\n\r\nNot included here (private): optimizer passes, scheduler, advanced kernels, full operator coverage, vendor-tuned implementations, and the full trust pipeline (content-addressed artifacts, Ed25519 signatures, attestations).\r\n\u27a1\ufe0f A full private demo is available under NDA.\r\n\r\n\ud83d\udd27 Useful commands\r\nList OpenCL devices (if PyOpenCL + drivers are present)\r\n\r\naxiomos-devices\r\n# universal fallback:\r\npython -m axiomos.cli.devices\r\nVerify a fixture (example: Softmax 8\u00d78)\r\npython -m axiomos.verify examples/softmax2d_small.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0\r\nOutput includes: SHAPES, max_abs_err, ALLCLOSE (atol=1e-6, rtol=1e-5), CPU & OPENCL timings.\r\nNote: the public OpenCL path is minimal (not optimized) \u2014 correctness, not performance.\r\n\r\nTiny smoke (latency indicator)\r\naxiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0\r\n# fallback:\r\npython -m axiomos.smoke --size 512 --warmup 3 --repeat 30 --seed 0\r\n\r\n\ud83e\uddea Optional: PyTorch \u2192 AXIR mini-export + verify\r\nInstall PyTorch (it\u2019s large; keep optional for demos):\r\npip install torch\r\nExport a tiny model to AXIR JSON:\r\naxiomos-export-torch\r\n# fallback:\r\n\r\nThis creates:\r\nexamples/pytorch_softmax.axir.json\r\nVerify CPU \u2194 OPENCL on the exported AXIR:\r\n\r\npython -m axiomos.verify examples/pytorch_softmax.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0\r\nAs with other demos: if OpenCL isn\u2019t available, verification falls back to OPENCL(cpu-fallback) and still checks numeric parity.\r\n\r\n\ud83e\uddf0 If you cloned the repo (extra fixtures)\r\nGenerate fixtures:\r\n\r\npython make_fixtures.py\r\nCreates:\r\nexamples/\r\n \u251c\u2500 vector_add_small.axir.json\r\n \u2514\u2500 softmax2d_small.axir.json\r\nVerify:\r\n\r\n# CPU \u2194 CPU (Hello AXIR)\r\naxiomos-verify examples/vector_add_small.axir.json --buffer hC --backend-a cpu --backend-b cpu --seed 0\r\n\r\n# CPU \u2194 OpenCL (Softmax 8\u00d78)\r\npython -m axiomos.verify examples/softmax2d_small.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0\r\n\r\n\ud83d\udca1 Common errors & fixes\r\n\u201cCommand not found\u201d (e.g., axiomos-devices):\r\nUse module form (works everywhere):\r\npython -m axiomos.cli.devices (Windows) or python3 -m axiomos.cli.devices (macOS/Linux).\r\nOn Windows, if using a venv, activate it before running commands.\r\n\r\n\u201cNo matching distribution found for axiomos\u201d\r\nPython is likely too old. Use Python 3.10+. On macOS:\r\nbrew install python@3.11\r\n/opt/homebrew/bin/python3.11 -m venv .venv   # Apple Silicon\r\n# or: /usr/local/bin/python3.11 -m venv .venv # Intel\r\nsource .venv/bin/activate\r\npython -m pip install -U pip setuptools wheel\r\npython -m pip install -U \"axiomos>=0.1.15\"\r\n\r\nApple Silicon arch mismatch (Rosetta vs arm64):\r\nusr/bin/arch -arm64 python3 -m venv .venv\r\nsource .venv/bin/activate\r\npython -m pip install -U pip\r\npython -m pip install -U \"axiomos>=0.1.15\"\r\nCorporate proxy / SSL errors:\r\nConfigure pip to use your proxy or trusted certs, or install inside a VPN/approved network. (Keep the module form.)\r\n\r\nPyTorch install is slow:\r\nThat\u2019s expected (large wheels). Keep PyTorch optional for the demo.\r\n\r\nWhy AXIOMOS\r\nAXIR (Axiomos IR) is a universal, hardware-agnostic IR aimed at:\r\n\r\nPortability \u2014 compile once, run on CPU, GPU, and accelerators.\r\n\r\nDeterminism & Reproducibility \u2014 numerically verifiable parity across backends.\r\n\r\nTrust \u2014 measured parity today; cryptographic provenance/signatures in the private build.\r\n\r\nLongevity \u2014 a stable IR beyond today\u2019s frameworks and vendor APIs.\r\n\r\nThis public showcase is intentionally minimal: it proves AXIR JSON \u2192 multi-backend execution \u2192 verification without exposing the private core.\r\n\r\nLicense\r\nMIT (public showcase only). The full runtime remains proprietary.\r\n",
    "bugtrack_url": null,
    "license": "MIT",
    "summary": "A lightweight AXIR playground to run numerical kernels across CPU and OpenCL",
    "version": "0.1.18",
    "project_urls": {
        "Documentation": "https://github.com/Aidenkuro10/axiomos-vitrine#readme",
        "Homepage": "https://github.com/Aidenkuro10/axiomos-vitrine",
        "Repository": "https://github.com/Aidenkuro10/axiomos-vitrine"
    },
    "split_keywords": [
        "opencl",
        " kernels",
        " gemm",
        " axir",
        " numpy",
        " runtime",
        " ai"
    ],
    "urls": [
        {
            "comment_text": null,
            "digests": {
                "blake2b_256": "429023e2c96864011fc9539cf29bd11b6ea4249ea67ec4775d5578d9a742639f",
                "md5": "156fa72983d2d4ba2a99630bdc83c44c",
                "sha256": "6df841c2e0d87ac7901abd912b191337c5e6c9a00563814970bacde93d42875e"
            },
            "downloads": -1,
            "filename": "axiomos-0.1.18-py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "156fa72983d2d4ba2a99630bdc83c44c",
            "packagetype": "bdist_wheel",
            "python_version": "py3",
            "requires_python": ">=3.10",
            "size": 13414,
            "upload_time": "2025-10-13T23:00:46",
            "upload_time_iso_8601": "2025-10-13T23:00:46.197218Z",
            "url": "https://files.pythonhosted.org/packages/42/90/23e2c96864011fc9539cf29bd11b6ea4249ea67ec4775d5578d9a742639f/axiomos-0.1.18-py3-none-any.whl",
            "yanked": false,
            "yanked_reason": null
        },
        {
            "comment_text": null,
            "digests": {
                "blake2b_256": "6eec7f2e40973f7809a098ab86fef7e3b4576829d19b336aaf5c905a15d42d4a",
                "md5": "18ee25f1d3cae71f7c325d2d0bd3e1aa",
                "sha256": "af4005197bac7fa46e374d5ebe47b13a8df2a572a742c21825560fd41a40a289"
            },
            "downloads": -1,
            "filename": "axiomos-0.1.18.tar.gz",
            "has_sig": false,
            "md5_digest": "18ee25f1d3cae71f7c325d2d0bd3e1aa",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": ">=3.10",
            "size": 12753,
            "upload_time": "2025-10-13T23:00:47",
            "upload_time_iso_8601": "2025-10-13T23:00:47.564874Z",
            "url": "https://files.pythonhosted.org/packages/6e/ec/7f2e40973f7809a098ab86fef7e3b4576829d19b336aaf5c905a15d42d4a/axiomos-0.1.18.tar.gz",
            "yanked": false,
            "yanked_reason": null
        }
    ],
    "upload_time": "2025-10-13 23:00:47",
    "github": true,
    "gitlab": false,
    "bitbucket": false,
    "codeberg": false,
    "github_user": "Aidenkuro10",
    "github_project": "axiomos-vitrine#readme",
    "travis_ci": false,
    "coveralls": false,
    "github_actions": false,
    "lcname": "axiomos"
}
Axiomos
Elapsed time: 2.56347s