Update README.md

README.md

@@ -4,6 +4,14 @@ datasets:
 - AbstractPhil/geometric-vocab
 pipeline_tag: zero-shot-classification
 ---
+# Formulas have been purified
+
+The newest vit_zana_nano train has shown a very clean curve.
+
+I've begun training a much deeper zana dubbed vit_zana_shaper. This model has 32 layers deep with MLP ratio of 1 and 2 attention heads, resting at about 3.5 million params or so.
+
+Lets see how she fares.
+
 # I've had an epiphany. We don't NEED transformer layers in their current form.
 
 David's architecture already solved this need with high-efficiency multi-stage geometric mathematics.