[yuchenlichuck]

Summary

Moves the reproducible environment from PyTorch 2.10.0 / torchvision 0.25.0 to PyTorch 2.11.0 / torchvision 0.26.0.

• requirements.txt: torch==2.11.0+cu128 / torchvision==0.26.0+cu128 on Linux and Windows, native torch==2.11.0 / torchvision==0.26.0 wheels on macOS. The cu128 extra index is unchanged; 2.11.0 ships cp312 cu128 wheels for manylinux_2_28_x86_64, manylinux_2_28_aarch64 and win_amd64, and PyPI has the macosx_11_0_arm64 wheels for both packages.
• README.md: Installation text updated to match.
• pyproject.toml: unchanged on purpose. The published-package ranges torch>=2.10,<2.12 / torchvision>=0.25,<0.27 already admit 2.11.0 / 0.26.0, following the split from 4aefe82 (library keeps ranges, env files carry the exact pins).
• deeplens/psfnetlens.py: the >= 2.10 floor for safe weights_only checkpoint loading is unaffected and still correct.

No source changes were needed; the library runs on 2.11 as-is.

Verification

Fresh Python 3.12.12 venv, python -m pip install -r requirements.txt, then pip install -e . --no-deps:

• torch 2.11.0+cu128, torchvision 0.26.0+cu128, CUDA 12.8 reported by torch.version.cuda
• pytest test/: 672 passed, 0 failed, 0 skipped, no warnings (RTX 6000 Ada, 88 s)
• 0_hello_geolens.py, 0_hello_defocuslens.py, 0_hello_diffraclens.py, 0_hello_hybridlens.py all run to completion on GPU
• ruff check . and ruff format --check . clean at ruff 0.16.5

🤖 Generated with Claude Code

[98zc5g5jyw-arch]

Verified against head 4375a9cabd493f78d609dbb25ce2e58dd15366ae (PyTorch 2.11.0 / torchvision 0.26.0 bump).

• ✅ Scope: exactly 2 files (requirements.txt, README.md, +10/−6) — nothing out of scope touched.
• ✅ Pin/README consistency: head requirements.txt has torch==2.11.0+cu128 / torchvision==0.26.0+cu128 for Linux/Windows and native torch==2.11.0 / torchvision==0.26.0 for Apple-silicon macOS; the README paragraph states the same. Base→head diff is only those 4 pins.
• ✅ Versions exist: torch 2.11.0 and torchvision 0.26.0 are on PyPI with macosx_11_0_arm64 wheels, and torchvision 0.26.0 declares torch==2.11.0 — the two move in lockstep.
• ✅ cu128 index spot-check: torch-2.11.0+cu128 listed for cp310–cp313 (manylinux_2_28 x86_64/aarch64, win_amd64, 21 entries); torchvision-0.26.0+cu128 cp312 manylinux_2_28_x86_64 and win_amd64 wheels both respond 200.
• ✅ Out-of-scope files untouched: pyproject.toml ranges (torch>=2.10,<2.12, torchvision>=0.25,<0.27) already admit 2.11.0/0.26.0 and intentionally keep 2.10 supported, consistent with the runtime "PyTorch 2.10 or newer" gate; repo-wide search shows cu128 / torch== only in README + requirements.
• ✅ The new Volta note matches upstream: PyTorch 2.11 release notes confirm CUDA 12.8/12.9 binaries drop sm_70, with cu126 as the fallback for V100-class GPUs.
• ⚪ CI: gh pr checks reports no checks on chore/torch-2.11, so wheel installability is not exercised by CI on this PR.

Suggestions (optional):

1. A small pip install -r requirements.txt smoke job on Linux + macOS runners would catch wheel-availability regressions on future bumps like this one.
2. The older 2.10.0 string in test/test_performance_report_regressions.py is an intentional fixture for the version-gate test, not a stale pin — no change needed, just flagging it so it isn't mistaken for one during a future sweep.