Search results

feat: improve gradient stability for extreme-scale inputs in half-precision

enhancementNew feature or requestNew feature or request

Status: Open.

#89 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 8, 2026

feat: chunked/streaming cross-covariance accumulation for large N

enhancementNew feature or requestNew feature or request

moderateModerate impact, fix when possibleModerate impact, fix when possible

performanceRuntime performance improvementRuntime performance improvement

Status: Open.

#29 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

docs: multi-GPU and distributed usage guide

docsDocumentation improvementDocumentation improvement

enhancementNew feature or requestNew feature or request

lowLow impact, nice to haveLow impact, nice to have

Status: Open.

#26 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

profile: JAX call overhead with and without jax.jit wrapping

benchmarkPerformance measurement or profilingPerformance measurement or profiling

framework:jaxJAX-specific issueJAX-specific issue

moderateModerate impact, fix when possibleModerate impact, fix when possible

performanceRuntime performance improvementRuntime performance improvement

Status: Open.

#25 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

profile: GPU occupancy of batched SVD on small D matrices

benchmarkPerformance measurement or profilingPerformance measurement or profiling

framework:pytorchPyTorch-specific issuePyTorch-specific issue

moderateModerate impact, fix when possibleModerate impact, fix when possible

performanceRuntime performance improvementRuntime performance improvement

Status: Open.

#24 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

benchmark: Horn vs. Kabsch throughput for 3D inputs

benchmarkPerformance measurement or profilingPerformance measurement or profiling

lowLow impact, nice to haveLow impact, nice to have

Status: Open.

#23 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

benchmark: quantify MLX CPU round-trip overhead from stream=mx.cpu SVD

benchmarkPerformance measurement or profilingPerformance measurement or profiling

framework:mlxMLX-specific issueMLX-specific issue

moderateModerate impact, fix when possibleModerate impact, fix when possible

performanceRuntime performance improvementRuntime performance improvement

Status: Open.

#22 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

benchmark: throughput vs. N (points per pair) across all frameworks

benchmarkPerformance measurement or profilingPerformance measurement or profiling

moderateModerate impact, fix when possibleModerate impact, fix when possible

Status: Open.

#21 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026

benchmark: throughput vs. batch size (B) across all frameworks

benchmarkPerformance measurement or profilingPerformance measurement or profiling

moderateModerate impact, fix when possibleModerate impact, fix when possible

Status: Open.

#20 In hunter-heidenreich/Kabsch-Cookbook;

· hunter-heidenreich opened on Mar 6, 2026