Upload README.md with huggingface_hub

README.md

@@ -18,10 +18,13 @@ packed signed INT4 weights with group-256 regular-Hadamard **ConvRot** rotation
 (arXiv [2512.03673](https://arxiv.org/abs/2512.03673)), activations rotated online
 and dynamically quantized inside the kernel. Calibration-free.
 
-Mixed recipe: 224 block Linears = **140 INT4 + 84 int8_tensorwise fallback**
-(`attn.wv`/`attn.wo` — the ConvRot paper's sensitive projections — plus
-`attn.gate`, modulation-class). The shared modulation projector `tproj` and the
-text-fusion transformer stay bf16.
+Mixed recipe: 224 block Linears = **128 INT4 + 96 int8_tensorwise fallback**,
+selected by measured per-layer sensitivity: each layer was swapped alone to
+W4A4 on real sampling inputs, ranked by final-output drift, then chosen by
+impact-per-byte at a fixed size budget (all `wv`/`wk`, 27/28 `wo`, the most
+sensitive `mlp.down`/`gate`/`wq` layers). +2.2 dB over the class-heuristic
+recipe at identical size and speed. The shared modulation projector `tproj`
+and the text-fusion transformer stay bf16.
 
 ## Samples
 
@@ -48,7 +51,7 @@ fails with a clear "format not available" error.
 | bf16 | 252 s |
 
 Weights: **9.25 GB** vs 28 GB bf16 (−67%) — fits a 24 GB card with headroom where
-bf16 needs weight streaming. PSNR vs bf16 (same seed): 22.1 dB (min 19.6, max 24.2).
+bf16 needs weight streaming. PSNR vs bf16 (same seed): 24.3 dB (min 18.6, max 31.0).
 
 ## Reproduce