Update README.md

README.md

@@ -19,7 +19,7 @@ datasets:
 
 # BinaryLLM (Proof of Concept)
 
-Tokenizer-free / base-2 (vocab_size=4) proof of concept.
+Tokenizer-free / radix-2 (vocab_size=4) proof of concept.
 
 This repo uses `trust_remote_code=True` because it ships a custom `modeling_*.py` / `configuration_*.py`.
 
@@ -88,7 +88,7 @@ Avi
 
 This repo includes a minimal inference script that:
 
-* encodes the prompt to base-2 bits (UTF-8, MSB→LSB),
+* encodes the prompt to radix-2 bits (UTF-8, MSB→LSB),
 * runs a manual token-by-token loop (no `generate`),
 * decodes the generated bits back to text (best-effort strict decode).