Training

Bytes Are All You Need

Chat with Paper

First page

Paper summary

Performs classification directly on file bytes without decoding.

Ask this paper

Key points

Raw-byte input: Trains Transformers directly on raw file bytes (PNG, WAV, etc.) rather than decoded tensors.

Strong results: Achieves 77.33% ImageNet Top-1 accuracy on raw bytes and 95.42% on raw WAV for Speech Commands v2.

Format-agnostic: A single architecture handles any file format without preprocessing pipelines.

Infrastructure simplification: Suggests a future where models eat raw bytes and skip format-specific codecs - simpler pipelines with less preprocessing error.

Every Monday

Bytes Are All You Need

Ask this paper

Get next week’s papers.