HHLO is a Haskell library and runtime for building, compiling, and executing machine-learning programs targeting StableHLO, the portable intermediate representation of the OpenXLA ecosystem. . Instead of replicating JAX's Python-based tracing infrastructure, HHLO generates StableHLO MLIR text directly from Haskell and compiles it to CPU or GPU via the PJRT plugin interface. . Key features: . * Type-safe EDSL with phantom-shape tensors (e.g. Tensor '[2,3] 'F32) * 50+ ops: matmul, conv2d, softmax, batch-norm, control flow, and more * CPU execution via PJRT CPU plugin * GPU execution via PJRT CUDA plugin with device enumeration * Multi-GPU concurrent inference scaling * No LLVM/MLIR build dependency — emits text and lets PJRT parse it . See the README for setup instructions and examples.