rlx-diamond 0.2.5

Diamond Maps reward alignment — flow matching value functions and GLASS sampling (arXiv:2602.05993)
Documentation
  • Coverage
  • 72.37%
    55 out of 76 items documented0 out of 62 items with examples
  • Size
  • Source code size: 33.17 kB This is the summed size of all the files inside the crates.io package for this release.
  • Documentation size: 836.96 kB This is the summed size of all files generated by rustdoc for all configured targets
  • Ø build duration
  • this release: 2s Average build duration of successful builds.
  • all releases: 3s Average build duration of successful builds in releases after 2024-10-23.
  • Links
  • MIT-RLX/rlx-models
    3 0 0
  • crates.io
  • Dependencies
  • Versions
  • Owners
  • eugenehp

Diamond Maps: inference-time reward alignment via stochastic flow maps.

Implements host-side math from arXiv:2602.05993 (value functions, GLASS flows, weighted renoising). Model integration lives in rlx-flux2::diamond.

No retraining of the base generative model is required for GLASS-based guidance; multi-step GLASS posterior sampling uses the frozen denoiser as a reference.