HXQ: calibration-free tensor codec. Compresses any dense tensor — model weights, embeddings, state vectors — with per-group affine quantization. No calibration data. Per-tensor receipts. MIT.
-
Updated
May 22, 2026 - C
#
tensor-codec
Here is 1 public repository matching this topic...
Add this topic to your repo
To associate your repository with the tensor-codec topic, visit your repo's landing page and select "manage topics."