Upload README.md with huggingface_hub

README.md

@@ -11,10 +11,11 @@ tags:
 
 # System One Model (using Gemma 4 12B)
 
-`google/gemma-4-12B` with its 262k-token language-model head replaced by a
-low dimentional vector that outputs one logit per answer. The decoder is
-unchanged. One forward pass gives the 26 logits; questions (`noul`, `choice`,
-`score`) are answered by softmax over the logits each question.
+Using encoder and decoder from `google/gemma-4-12B` with its 262k-token
+language-model head replaced by a low dimentional vector
+that outputs one logit per answer. One forward pass gives the 26 logits;
+questions (`noul`, `choice`, `score`) are answered by softmax
+over the logits each question.
 
 Requires `transformers>=5.17`. The repo ships custom code, so pass
 `trust_remote_code=True`.