GitHub user 400Ping added a comment to the discussion: [DISCUSSION] Apache Mahout Roadmap: Q3–Q4 2026
Thanks for putting this together @rich7420. The proposal looks good to me, and I think the five areas capture the current pain points well (especially separating the benchmark tiers and the fail-fast VRAM validation work). One direction I'd like to add to the roadmap: distributed QDP, or more generally a distributed execution story for Mahout's quantum stack. Motivation: - QDP currently scales within a single device (batching, async prefetch, memory planning), but state preparation over large datasets is embarrassingly parallel across samples, sharding encode/prepare work across multiple GPUs or nodes is a natural next step. - The memory-estimation and VRAM-validation work already on the roadmap gives us exactly the primitives a distributed planner would need to partition work by device capability. - It also fits Mahout's history as a distributed ML project, a "distributed quantum data preparation" layer would be a differentiator that PennyLane/Qiskit don't really offer today. Scope-wise I'd suggest starting small: multi-GPU single-node sharding of QDP batches first, then evaluate whether a multi-node layer (e.g. over Ray/Dask or plain MPI-style launchers) is worth prototyping. GitHub link: https://github.com/apache/mahout/discussions/1465#discussioncomment-18174420 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
