Paper summary
Paper summary: The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
BitNet b1.58 restricts every weight to −1, 0 or +1 (about 1.58 bits) and reports matching full-precision LLMs of the same size and data on perplexity and end tasks, at much lower cost.
- Adding 0 to binary weights gives ternary {−1, 0, 1}: log₂3 ≈ 1.58 bits per weight.
- The authors argue this defines a new scaling law and recipe, and motivates hardware built for 1-bit LLMs.