-
Notifications
You must be signed in to change notification settings - Fork 0
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
feat: improve gradient stability for extreme-scale inputs in half-precision
enhancementNew feature or requestNew feature or requestStatus: Open.#89 In hunter-heidenreich/Kabsch-Cookbook;feat: chunked/streaming cross-covariance accumulation for large N
enhancementNew feature or requestNew feature or requestmoderateModerate impact, fix when possibleModerate impact, fix when possibleperformanceRuntime performance improvementRuntime performance improvementStatus: Open.#29 In hunter-heidenreich/Kabsch-Cookbook;docs: multi-GPU and distributed usage guide
docsDocumentation improvementDocumentation improvementenhancementNew feature or requestNew feature or requestlowLow impact, nice to haveLow impact, nice to haveStatus: Open.#26 In hunter-heidenreich/Kabsch-Cookbook;profile: JAX call overhead with and without jax.jit wrapping
benchmarkPerformance measurement or profilingPerformance measurement or profilingframework:jaxJAX-specific issueJAX-specific issuemoderateModerate impact, fix when possibleModerate impact, fix when possibleperformanceRuntime performance improvementRuntime performance improvementStatus: Open.#25 In hunter-heidenreich/Kabsch-Cookbook;profile: GPU occupancy of batched SVD on small D matrices
benchmarkPerformance measurement or profilingPerformance measurement or profilingframework:pytorchPyTorch-specific issuePyTorch-specific issuemoderateModerate impact, fix when possibleModerate impact, fix when possibleperformanceRuntime performance improvementRuntime performance improvementStatus: Open.#24 In hunter-heidenreich/Kabsch-Cookbook;benchmark: Horn vs. Kabsch throughput for 3D inputs
benchmarkPerformance measurement or profilingPerformance measurement or profilinglowLow impact, nice to haveLow impact, nice to haveStatus: Open.#23 In hunter-heidenreich/Kabsch-Cookbook;benchmark: quantify MLX CPU round-trip overhead from stream=mx.cpu SVD
benchmarkPerformance measurement or profilingPerformance measurement or profilingframework:mlxMLX-specific issueMLX-specific issuemoderateModerate impact, fix when possibleModerate impact, fix when possibleperformanceRuntime performance improvementRuntime performance improvementStatus: Open.#22 In hunter-heidenreich/Kabsch-Cookbook;benchmark: throughput vs. N (points per pair) across all frameworks
benchmarkPerformance measurement or profilingPerformance measurement or profilingmoderateModerate impact, fix when possibleModerate impact, fix when possibleStatus: Open.#21 In hunter-heidenreich/Kabsch-Cookbook;benchmark: throughput vs. batch size (B) across all frameworks
benchmarkPerformance measurement or profilingPerformance measurement or profilingmoderateModerate impact, fix when possibleModerate impact, fix when possibleStatus: Open.#20 In hunter-heidenreich/Kabsch-Cookbook;