Colendi AIResearch
Menu

Transformer Architecture · 2024

Byte Latent Transformer: Patches Scale Better Than Tokens (BLT)

Artidoro Pagnoni, Ram Pasunuru, Pedro Rodriguez, et al. (Meta FAIR)

Primary source

Removes the fixed tokenizer by grouping raw bytes into entropy-based patches, matching Llama-3 quality with up to 50% fewer inference FLOPs and better robustness.

Editorial record

Plain-language summary

Removes the fixed tokenizer by grouping raw bytes into entropy-based patches, matching Llama-3 quality with up to 50% fewer inference FLOPs and better robustness.

Knowledge graph

Relationships

Descendants

  • Depends onEvidence: Strongly supported

    Attention Is All You Need

    Builds on the transformer architecture lineage in the archive

    freshness sweep 2026

Source record

Provenance

Record ID
P-612
Record created
2026-07-13
Last reviewed
2026-07-14
Record version
2

Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.