DEPLOY.md

Deployment Guide

Two targets, two strategies (GitHub's free LFS tier is only 1GB, so the 8.3GB tablebase goes to Hugging Face only):

Target	Contents
GitHub	Code only — layers/ stays gitignored
HF Space	Code + Dockerfile + layers/compressed (8.3GB, LFS)

1. GitHub (code-only)

git add -A
git commit -m "Restructure: promote New/ to root, add Tablebase Explorer"
git remote add origin git@github.com:<you>/Bestemshe.git
git push -u origin audit-optimize-harden   # or merge to main first

Nothing under layers/ is committed (see .gitignore), so no LFS needed on GitHub.

2. Hugging Face Space

The Space ansarzeinulla/Bestemshe-God-Algorithm already exists as a free Gradio SDK space (Docker Spaces require a PRO subscription; Gradio ones are free). The C++ CLI is compiled at app startup — packages.txt installs g++/make/libzstd-dev via apt, and app.py builds ./query on first launch. Do NOT upload the Dockerfile or a README with sdk: docker — that converts the Space to Docker and triggers a 402.

pip install -U "huggingface_hub[cli]"
hf auth login    # the CLI is `hf` (huggingface-cli is deprecated)

Option A (recommended): HfApi.upload_folder

NOTE: hf upload fails with 402 Payment Required on free-tier Gradio Spaces because the CLI always hits the repos/create endpoint first (even for existing repos). The Python API commits directly to the existing repo and works. Run from the repo root:

from huggingface_hub import HfApi
api = HfApi()

# Code
api.upload_folder(
    repo_id="ansarzeinulla/Bestemshe-God-Algorithm", repo_type="space",
    folder_path=".",
    allow_patterns=["app.py", "requirements.txt", "packages.txt", "query.cpp",
                    "Oracle.h", "StateIndex.h", "BestemsheCore.h", ".gitattributes", "README.md"],
)

# Tablebase (8.3GB, LFS automatically)
api.upload_folder(
    repo_id="ansarzeinulla/Bestemshe-God-Algorithm", repo_type="space",
    folder_path="layers/compressed", path_in_repo="layers/compressed",
)

This uploads code + layers/compressed (large files automatically stored as LFS server-side). Resumable if the 8.3GB transfer drops. The Space then builds the Dockerfile and serves the Gradio app on port 7860.

Option B: pure git-lfs push

git clone https://huggingface.co/spaces/ansarzeinulla/Bestemshe-God-Algorithm hf-space
cd hf-space
git lfs install
cp -R ../{app.py,requirements.txt,Dockerfile,README.md,.gitattributes,query.cpp,Oracle.h,StateIndex.h,BestemsheCore.h} .
mkdir -p layers && cp -R ../layers/compressed layers/compressed
# IMPORTANT: do NOT copy the repo .gitignore into the HF clone — it ignores layers/.
git add -A
git commit -m "Bestemshe Tablebase Explorer"
git push   # pushes 8.3GB via LFS; expect it to take a while

.gitattributes already routes layers/compressed/*.bin through LFS.

Notes

• The Docker image COPYs the tablebase into the image; first build takes a while but queries never load more than one 4MB block into RAM.
• To test the container locally: docker build -t bestemshe . && docker run -p 7860:7860 bestemshe.