Fix primary codebook penalty

tools/llama/generate.py

@@ -97,7 +97,7 @@ def decode_one_token(
     codebooks = [
         sample(
             logits.token_logits,
-            previous_tokens=previous_tokens[0],
+            previous_tokens=None,  # Disable repetition penalty for the token codebook
             **sampling_kwargs,
         )[0]
     ]