// ◈ Quantization, Pruning & Distillation
BitNet b1.58 (Ternary LLMs)
Trains LLMs from scratch with ternary weights {−1, 0, +1}.
BitNet b1.58 三值模型 2024 Advanced core
IN PLAIN WORDS
With ternary weights matrix multiplication becomes additions, slashing memory, energy and latency. Matches FP16 at 3B scale, but must be trained from scratch.
Used in: bitnet.cpp, edge inference research
01Original paper
The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
Shuming Ma, Hongyu Wang, Lingxiao Ma, et al. · 2024
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Learn first
BitNet b1.58 (Ternary LLMs)
Then you can learn
- Nothing in the atlas builds on it yet
Related:GPTQ
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
Reference code
04Where the course teaches it
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
bitnet)