The latest built artifact for x86 linux appears to be on June 7th. The build scheduled on June 8th failed with a pip error mentioning mesonpy, and that same error still occurs for every build. This coincides with the release of numpy 2.3.0, and indeed, the github actions logs for the builds on June 7th show that numpy 2.2.6 was downloaded, while the run on June 8th showed that numpy 2.3.0 was downloaded.

The CI builds on the main repo still succeed, and looking through their logs it seems that the CI they use installs the dependencies for MLIR's python bindings (https://github.com/llvm/torch-mlir/blob/46c3888a3b861473ea9bb90f6033b5fa41a0ca6c/build_tools/ci/install_python_deps.sh#L10), which specifies the numpy version to be between 1.19.5 and 2.1.2 (https://github.com/llvm/llvm-project/blob/38318dd05615a2f38abdeeae99e7423165308902/mlir/python/requirements.txt#L2C8-L2C14). However, pytorch_mlir itself sets no restrictions on the numpy version, so it pulls the latest available when it has not yet been installed.