Update README.md

README.md

@@ -9,12 +9,31 @@ tags: [qwen3, byte-level, pretraining]
 
 # details
 
-state of the art byte-level language model pretrained from scratch on rewrite, climbmix, openmath, ultrafineweb and cortex (our internal symbolic kernel).
-
-90,148,352 tied-embedding parameters. 
+State of the art, small language model pretrained from scratch on rewrite, climbmix, openmath, ultrafineweb and cortex (our internal symbolic kernel).
 
 259 tokens: bytes 0–255, PAD 256, BOS 257, EOS 258.
 
-we used revision `92c601560cc2179e3bfc33e067fdd3bf9254e163` (step 900,000) for evaluations.
+## Benchmark Results
+
+Model: `appvoid/void-byte` · Checkpoint: Step 900,000
+
+| Benchmark                    | Score   |
+| ---------------------------- | ------- |
+| HellaSwag                    | 38.68%  |
+| PIQA                         | 67.46%  |
+| ARC-Easy                     | 47.31%  |
+| ARC-Challenge                | 28.16%  |
+| ArithMark-3                  | 44.80%  |
+| Intelligence Index           | 23.9182 |
+| Benchmark Max Length         | 2,048   |
+
+
+| Property     | Value                                                         |
+| ------------ | ------------------------------------------------------------- |
+| Parameters   | 90,148,352 - tied-embedding                                                           |
+| Architecture | Byte-level                                                    |
+| Revision     | `92c601560cc2179e3bfc33e067fdd3bf9254e163`                    |
+
+We used the revision on step 900,000 for evaluations which trained for around 120 billion bytes which is around 30 to 35 billion bpe tokens.
 
-for more details on the evals, please have a look at the official notebook.
+For more details on the evals, please have a look at the official notebook.