# Format and quant references
llama.cpp                  commit 5e9f3a2 (2026-01)    github.com/ggerganov/llama.cpp
gguf-py                    0.19.0                       github.com/ggerganov/llama.cpp/tree/master/gguf-py

# Entropy coders
constriction               0.4                          github.com/bamler-lab/constriction
zstandard (Python binding) 0.22                         github.com/indygreg/python-zstandard

# Model loading
safetensors                0.4.5                        github.com/huggingface/safetensors

# Clustering / analysis
scikit-learn               1.4                          scikit-learn.org

# Trained-profile codec
openzl                     0.4.1                        github.com/facebook/openzl

# Comparison rows (no reproductions vendored in v4.0.3; see DEVIATIONS.md §C)
DFloat11 (official)        —                            github.com/LeanModels/DFloat11
                                                        Zhang et al., arXiv:2504.11651, NeurIPS 2025
                                                        OpenReview: https://openreview.net/forum?id=xdNAVP7TGy
Cloudflare Unweight        —                            github.com/cloudflareresearch/unweight-kernels  (CUDA, Hopper-only)
                                                        Nikulin, Cloudflare Research, 2026
                                                        Tech report: https://research.cloudflare.com/nikulin2026/
                                                        Blog: https://blog.cloudflare.com/unweight-tensor-compression/
ZipNN                      —                            github.com/zipnn/zipnn
                                                        Hershcovitch et al., arXiv:2411.05239

# Hardware reference
host                       c3-highcpu-88, Intel Sapphire Rapids, single-core
os                         Debian 12
python                     3.11