# REPRODUCTION NOTICE — `dfloat11`

This directory is reserved for a faithful re-implementation of **DFloat11** from its published description (not the original authors' code).

## Status: NOT IMPLEMENTED in this artifact

Official implementation: **https://github.com/LeanModels/DFloat11**

DFloat11 is described in: Long et al., *"70% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float"*, arXiv:2504.11651.
OpenReview: https://openreview.net/forum?id=xdNAVP7TGy

See `../../DEVIATIONS.md §C` for the gap and recovery plan. Until either the internal re-implementation lands here under spec §1.4 reproduction headers or the official kernels are run on the bf16 test corpus, paper Table 1's "1.474× re-implementation estimate" row should be scoped as "expected from published description, not vendored in this artifact."