Add one-time v0.11.0 release helper

Author: lilv98

.github/release-notes-v0.11.0.md

@@ -0,0 +1,29 @@
+## Maia-2 0.11.0
+
+This release focuses on reproducible training, safer checkpoint recovery, and maintained Rapid/Blitz training workflows. The released Rapid and Blitz model weights remain unchanged.
+
+### Highlights
+
+- Added maintained Rapid and Blitz training presets with strict, speed-specific Lichess filtering and separate checkpoint roots.
+- Reworked monthly training artifacts around validated JSON chunk caches, cryptographic data provenance, atomic downloads, decompression, and checkpoints, with clear preprocessing-worker failures.
+- Added resumable checkpoints that record configuration, source data, optimizer progress, and RNG state, and reject incompatible or incomplete resumes before loading optimizer state.
+- Improved indexed CUDA and multi-GPU handling, plus portable optimizer-state restoration across CPU, CUDA, and Apple MPS.
+- Tightened inference and evaluation behavior for terminal positions, legal-move masking, and prediction ties after display rounding.
+- Added checksum verification for released models and example datasets, a configurable Lichess download script, packaged training configurations, Python 3.10–3.12 CI, CodeQL, and expanded project metadata and documentation.
+
+### Compatibility notes
+
+- Python 3.10–3.12 and PyTorch 2.8.x are supported (`torch>=2.8,<2.9`).
+- Standard inference APIs and the released Rapid/Blitz weights are unchanged.
+- Training scripts must use an `if __name__ == "__main__":` guard because preprocessing uses spawned workers.
+- Configurations without `game_type` continue to default to Rapid. Legacy checkpoints without training metadata can be resumed only as Rapid.
+- Legacy `*_chunks.pkl` caches are rebuilt once as provenance-bound JSON caches.
+- `from_pretrained` now uses the packaged model configuration rather than a mutable `save_root/config.yaml`.
+- Training requires case-sensitive `Rated` plus the selected `Rapid` or `Blitz` Event marker; games involving BOT titles or missing per-move clocks are excluded.
+
+### Distribution checksums
+
+- `maia2-0.11.0-py3-none-any.whl`: `939e69587026be62cd003b8cd1e6e2e66f1120390f68b9d363ce3da099c85029`
+- `maia2-0.11.0.tar.gz`: `d7c02445fb5d24e9beb6a19aaf3a5beb96192847424faa1cb120a9471385072f`
+
+**Full Changelog**: https://github.com/CSSLab/maia2/compare/v0.10.0...v0.11.0

.github/workflows/publish-v0.11.0-release.yml

@@ -0,0 +1,42 @@
+name: Publish v0.11.0 GitHub Release
+
+on:
+  push:
+    branches:
+      - release-automation-v0.11.0
+
+permissions:
+  contents: write
+
+jobs:
+  release:
+    if: github.repository == 'CSSLab/maia2'
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - name: Check out release helper
+        uses: actions/checkout@v7
+
+      - name: Download published distributions
+        run: |
+          curl -fL -o maia2-0.11.0-py3-none-any.whl \
+            https://files.pythonhosted.org/packages/9c/21/6ff621782cde9a98878d87e2a989a875f5a93f5abf207ccb8fc5acde3224/maia2-0.11.0-py3-none-any.whl
+          curl -fL -o maia2-0.11.0.tar.gz \
+            https://files.pythonhosted.org/packages/87/b3/03a6fc05eef1a01f3572e40ba8fbdd2e96dfb3976325f5b9215ffdf7792e/maia2-0.11.0.tar.gz
+
+      - name: Verify distribution checksums
+        run: |
+          echo '939e69587026be62cd003b8cd1e6e2e66f1120390f68b9d363ce3da099c85029  maia2-0.11.0-py3-none-any.whl' | sha256sum -c -
+          echo 'd7c02445fb5d24e9beb6a19aaf3a5beb96192847424faa1cb120a9471385072f  maia2-0.11.0.tar.gz' | sha256sum -c -
+
+      - name: Create GitHub Release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          gh release create v0.11.0 \
+            'maia2-0.11.0-py3-none-any.whl#Python wheel' \
+            'maia2-0.11.0.tar.gz#Source distribution' \
+            --repo "$GITHUB_REPOSITORY" \
+            --verify-tag \
+            --title 'Maia-2 0.11.0' \
+            --notes-file .github/release-notes-v0.11.0.md