Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models

TL;DR

Scratchpad Patching enhances byte-level language models by dynamically updating context within patches, reducing compute and cache needs while maintaining or improving model quality.

Contribution

The paper introduces Scratchpad Patching, a novel method that decouples compute from patch size by inserting transient scratchpads to improve model performance.

Findings

SP matches baseline quality at 16-byte patches with less compute.

SP reduces KV cache by 16 times and inference compute by 3-4 times.

SP improves natural language and code modeling performance.

Abstract

Tokenizer-free language models eliminate the tokenizer step of the language modeling pipeline by operating directly on bytes; patch-based variants further aggregate contiguous byte spans into patches for efficiency. However, the average patch size chosen at the model design stage governs a tight trade-off: larger patches reduce compute and KV-cache footprint, but degrade modeling quality. We trace this trade-off to patch lag: until a patch is fully observed, byte predictions within it must rely on a stale representation from the previous patch to preserve causality; this lag widens as patches grow larger. We introduce Scratchpad Patching (SP), which inserts transient scratchpads inside each patch to aggregate the bytes seen so far and refresh patch-level context for subsequent predictions. SP triggers scratchpads using next-byte prediction entropy, selectively allocating compute to information-dense regions and enabling post-hoc adjustment of inference-time compute. Across experiments on natural language and code, SP improves model quality at the same patch size; for example, even at $16$ bytes per patch, SP-augmented models match or closely approach the byte-level baseline on downstream evaluations while using a $16\times$ smaller KV cache over patches and $3$-$4\times$ less inference compute.