axiomos 0.1.12

A lightweight AXIR playground to run numerical kernels across CPU and OpenCL

AXIOMOS — Public Showcase (Minimal)

PyPI • GitHub

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

Why AXIOMOS (Vision)

Modern AI stacks are fragmented. AXIR (Axiomos IR) is a universal, hardware-agnostic IR for:

• Portability — compile once, run on CPU, GPU, accelerators.
• Determinism & Reproducibility — numerically verifiable parity across backends.
• Trust — measurable 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 revealing the private core.

---

Quickstart (works outside the repo)

1) Create a fresh env & install

Windows (PowerShell)

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install axiomos
# Optional: OpenCL support (requires vendor drivers)
pip install "axiomos[opencl]"

macOS / Linux

python3 -m venv .venv
source .venv/bin/activate
pip install axiomos
# Optional: OpenCL support (requires vendor drivers)
pip install "axiomos[opencl]"

2) Environment check
axiomos-doctor

3) Quick demo (auto-generates a fixture)
axiomos-demo

This creates examples/demo_vector_add.axir.json and runs a CPU ↔ OPENCL check.
If no OpenCL runtime is available, it falls back to OPENCL(cpu-fallback) and still verifies parity.

4) Optional tiny latency smoke test
axiomos-smoke --size 512 --warmup 3 --repeat 30 --seed 0

If you cloned the repo (extra fixtures)
python make_fixtures.py

Creates:

examples/
 ├─ vector_add_small.axir.json
 └─ softmax2d_small.axir.json

Verify examples (repo fixtures)
# 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
If pyopencl + drivers are present → label OPENCL(…) and the device path is used.

Otherwise → OPENCL(cpu-fallback).

Output includes: SHAPES, max_abs_err, ALLCLOSE (atol=1e-6, rtol=1e-5), CPU time, OPENCL time.
Note: OpenCL path = public minimal backend (not optimized).

(Optional) PyTorch → AXIR mini-demo

pip install torch
python pytorch_to_axir.py
python -m axiomos.verify examples/pytorch_softmax.axir.json --buffer hY --backend-a cpu --backend-b opencl --seed 0
What this proves (today)
Portability — the same .axir.json runs on CPU and OpenCL (if present).

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

Reproducibility — seedable runs and versioned fixtures.

Operable UX — install, run, verify in minutes; no private code exposure.

What’s intentionally not here
Private optimizer passes, schedulers, and advanced kernels.

Full operator coverage and vendor-tuned implementations.

Cryptographic provenance & signature pipeline (available in private build).

➡️ For the full demo (under NDA), please contact us.

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