Note NeurIPS Reject provenance in GRT-10M model card

README.md

@@ -5,6 +5,7 @@ tags:
   - geometric-rotation-transformer
   - causal-lm
   - rotational
+  - neurips-reject
 datasets:
   - nvidia/Nemotron-Pretraining-Dataset-sample
 language:
@@ -13,15 +14,17 @@ library_name: transformers
 pipeline_tag: text-generation
 ---
 
-# GRT-10M
+# GRT-10M (NeurIPS Reject Edition)
+
+> **Status: Official NeurIPS Reject.** Originally built as the small-scale proof-of-concept for our rejected NeurIPS submission introducing the Geometric Rotation Transformer (`GRT`). Kept intact as our historical `8.7M`-parameter (`2.2M` non-embedding backbone) baseline checkpoint—and now evaluated across the 10-task `SLM-Bench` suite with our classical symplectic & Huygens curriculum optimizers (`SymplecticAdam` and `HuygensSymplecticAdam`).
 
 A Geometric Rotation Transformer pretrained on Nemotron. GRT replaces the dense
 linear projections of a standard transformer with cascades of learned Givens
 rotations on SO(D), so the residual path contains no `nn.Linear` at all.
 
-This is a research proof-of-concept at small scale. It is published to
+This is a research proof-of-concept at small scale (and a proud NeurIPS reject). It is published to
 demonstrate that the architecture trains stably and evaluates end-to-end, not
-as a capable language model.
+as a capable general-purpose assistant.
 
 ## Model Details