BitTern aims to provide low-cost, high-accuracy post-training ternary quantization tools, as well as 1.58-bit models across diverse architectures, model scales, and reasoning tasks. Its goal is to lower the barrier to entry for developing 1.58-bit models, enabling broader community participation and allowing everyone can contribute and benefit from shared tools and models.
| Project | Venue | Public Release |
|---|---|---|
| CAT-Q | ICML 2026 Oral | Model checkpoints, inference, evaluation, and packed ternary deployment code |
[18/08/2026]🔥 Real ternary model deployment code is now available.[04/08/2026]🔥 The technical report of ScaleQ-1.58 "Attend to Your Own Thoughts: Breaking the Barrier for Post-Training Quantization of Reasoning LLMs through the Lens of 1.58-Bit Quantization" is now available on arXiv.[22/07/2026]🚀 The CAT-Q model checkpoints (including Qwen3-1.7B/4B/8B/14B/32B, Llama2-7B, Qwen3-30B-A3B and Qwen3-235B-A22B), inference, evaluation, and real ternary deployment code are now available.[25/06/2026]🔥 The CAT-Q paper is now available on arXiv.[01/05/2026]🎉Our paper "CAT-Q: Cost-efficient and Accurate Ternary Quantization for LLMs" is accepted to ICML 2026 as an oral. The project page for our sliding-layer reconstruction framework used in CAT-Q is available at SliderQuant (ICLR 2026).