docs: update README for pip install and add arXiv endorsement CTA

Author: SCAO Authors

MANIFEST.in

@@ -3,10 +3,13 @@ include README.md
 include CHANGELOG.md
 include pyproject.toml
 
-recursive-include scao *.py
+recursive-include scao *.py *.cu *.h
 recursive-include configs *.yaml
 exclude scao/benchmarks/*.py
 exclude scao/tests/*.py
 exclude scripts/*
 exclude paper/*
 exclude results_*.csv
+exclude report_*.txt
+exclude *.csv
+

README.md

@@ -8,7 +8,17 @@
 [![PyTorch](https://img.shields.io/badge/pytorch-2.0%2B-orange)](https://pytorch.org)
 
 > **A second-order PyTorch optimizer that delivers Shampoo-quality preconditioned gradients at near-AdamW memory and throughput cost.**  
-> Drop-in replacement for `AdamW`. One-line change. Real gains.
+> Drop-in replacement for `AdamW`. One-line change. Real gains.  
+> **Now available on PyPI:** `pip install scao`
+
+---
+
+## 🚀 Support the Research
+
+If you have endorsement rights on arXiv for **cs.LG** (Machine Learning), please consider endorsing our paper to help us share this work with the community:
+
+👉 **[Endorse SCAO on arXiv](https://arxiv.org/auth/endorse?x=X3VJ88)**
+
 
 ---
 
@@ -30,14 +40,15 @@
 
 ### Objection 3 — "It's lab code. Not suitable for the real world."
 
-**Test:** Eliminated every dependency on PyTorch or Hugging Face internals ([`examples/scao.py`](examples/scao.py))
+**Test:** SCAO is now a professional Python package, installable via PyPI ([`pip install scao`](https://pypi.org/project/scao/)).
 
-**Result:** SCAO is now a **single file** — a true drop-in replacement. Running natively on Windows with no cloud setup, the loss dropped from **4.536 → 3.307 in under 4 minutes**. The model learned real-world context: *"The secret to a good software architecture is its openness."*
+**Result:** SCAO has moved from a research script to a production-ready package. It's a true drop-in replacement for AdamW. Running natively on Windows with no cloud setup, the loss dropped from **4.536 → 3.307 in under 4 minutes**. The model learned real-world context: *"The secret to a good software architecture is its openness."*
 
 ---
 
 ## Table of Contents
 
+- [🚀 Support the Research](#-support-the-research)
 1. [The Problem](#1-the-problem)
 2. [SCAO's Solution](#2-scaos-solution)
 3. [Algorithm](#3-algorithm)
@@ -643,7 +654,6 @@ scao/                               # Core library
     └── setup.py                    # nvcc build (sm_70/75/80/86/89/90)
 
 examples/                           # Self-contained runnable examples
-├── scao.py                         # Standalone single-file SCAO (no library install needed)
 ├── train_local.py                  # Fine-tune GPT-2 125M with SCAO + LoRA (<8 GB VRAM)
 ├── train_1m.py                     # Full fine-tuning throughput benchmark on TinyStories-1M
 └── inference.py                    # Load LoRA checkpoint and generate text

examples/scao.py

@@ -66,7 +66,7 @@ all = [
 [tool.setuptools.packages.find]
 where = ["."]
 include = ["scao*"]
-exclude = ["scao/tests*", "scao/benchmarks*", "scao/cuda*"]
+exclude = ["scao/tests*", "scao/benchmarks*"]
 
 [tool.setuptools.package-data]
 scao = ["py.typed"]

pyproject.toml

@@ -0,0 +1,50 @@
+from setuptools import setup, find_packages
+
+setup(
+    name="scao",
+    version="0.1.1",
+    packages=find_packages(),
+    install_requires=[
+        "torch>=2.0.0",
+    ],
+    extras_require={
+        "dev": [
+            "pytest>=7.0",
+            "pytest-cov",
+            "mypy>=1.5",
+            "ruff>=0.1",
+            "build",
+            "twine",
+        ],
+        "cuda": [
+            "torch>=2.0.0",
+        ],
+        "hf": [
+            "transformers>=4.30.0",
+            "datasets>=2.0.0",
+        ],
+        "all": [
+            "transformers>=4.30.0",
+            "datasets>=2.0.0",
+            "mypy>=1.5",
+            "ruff>=0.1",
+        ],
+    },
+    author="SCAO Authors",
+    description="Sparse Curvature-Aware Adaptive Optimizer — second-order training at near-AdamW cost",
+    long_description=open("README.md", encoding="utf-8").read(),
+    long_description_content_type="text/markdown",
+    url="https://github.com/whispering3/scao",
+    classifiers=[
+        "Development Status :: 4 - Beta",
+        "Intended Audience :: Science/Research",
+        "Operating System :: OS Independent",
+        "Programming Language :: Python :: 3",
+        "Programming Language :: Python :: 3.10",
+        "Programming Language :: Python :: 3.11",
+        "Programming Language :: Python :: 3.12",
+        "Topic :: Scientific/Engineering :: Artificial Intelligence",
+        "Topic :: Software Development :: Libraries :: Python Modules",
+    ],
+    python_requires=">=3.10",
+)