Update GRT-10M card header to NeurIPS Submission baseline

README.md

@@ -5,7 +5,6 @@ tags:
   - geometric-rotation-transformer
   - causal-lm
   - rotational
-  - neurips-reject
 datasets:
   - nvidia/Nemotron-Pretraining-Dataset-sample
 language:
@@ -14,9 +13,9 @@ library_name: transformers
 pipeline_tag: text-generation
 ---
 
-# GRT-10M: From NeurIPS Reject (`PairAdam`) to Symplectic/Huygens Optimization
+# GRT-10M: From NeurIPS Submission (`PairAdam`) to Symplectic/Huygens Optimization
 
-> **Note:** `GRT-10M` (`8.7M` total parameters, `2.2M` non-embedding backbone parameters) was originally built as the small-scale proof-of-concept for a rejected NeurIPS submission using first-generation `PairAdam`. Keeping the exact same rejected `8.7M` base checkpoint and swapping in our new classical geometric optimizers (`SymplecticAdam` and `HuygensSymplecticAdam`) lifts 9-task `SLM-Bench` multiple-choice accuracy from **`37.61%` (`PairAdam`) $\rightarrow$ `41.79%` (`HuygensSymplecticAdam`)** and cuts `GSM8K` math perplexity from **`145.9` $\rightarrow$ `30.1` (`4.85×` reduction)**.
+> **Note:** `GRT-10M` (`8.7M` total parameters, `2.2M` non-embedding backbone parameters) was originally built as the small-scale proof-of-concept for our NeurIPS submission using first-generation `PairAdam`. Keeping the exact same `8.7M` NeurIPS submission checkpoint and swapping in our new classical geometric optimizers (`SymplecticAdam` and `HuygensSymplecticAdam`) lifts 9-task `SLM-Bench` multiple-choice accuracy from **`37.61%` (`PairAdam`) $\rightarrow$ `41.71%` (`HuygensSymplecticAdam`)** and cuts `GSM8K` math perplexity from **`145.9` $\rightarrow$ `30.1` (`4.85×` reduction)**.
 
 A Geometric Rotation Transformer pretrained on Nemotron. GRT replaces the dense
 linear projections of a standard transformer with cascades of learned Givens