First 1.58-bit open-source large model BitCPM-CANN fully trained on Huawei Ascend 910B NPU released
Original title:BitCPM-CANN just became the world’s first open-sourced 1.58-bit ternary LLM trained entirely on Chin...
AI summary
ModelBest, Tsinghua University, and the OpenBMB community jointly released BitCPM-CANN, the world's first open-source 1.58-bit ternary large model fully trained on Huawei Ascend 910B NPUs. Its core innovation is extremely low-bit quantization with only three weight states, reducing model memory usage by about 6x compared with BF16 and enabling efficient deployment on edge devices such as phones, computers, and in-vehicle devices. More importantly, the entire training stack—from quantization operators to frameworks—was natively built and validated on Ascend rather than simply ported. The model family (0.5B-8B) maintains 95-97% performance of full-precision models on multiple benchmarks, providing a practical solution for deploying and reproducing large models in resource-constrained environments.
Key Takeaways
- ModelBest,Tsinghua University,and the OpenBMB community jointly released BitCPM-CANN,the world's…
- 58-bit ternary large model fully trained on Huawei Ascend 910B NPUs.
- Its core innovation is extremely low-bit quantization with only three weight states,reducing model…
- More importantly,the entire training stack—from quantization operators to frameworks—was natively…
- The model family (0.
Quick Facts
- Published
- Updated
- Source
- X:Rohan Paul (@rohanpaul_ai)
- Category
- Model Releases
- AI score
- 75
首个开源的1.58-bit三元LLM,直接在昇腾芯片上原生训练,内存压缩到BF16的六分之一,8B模型就能跑在手机上,做端侧部署的可以立刻上手试试了。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: First 1.58-bit open-source large model BitCPM-CANN fully trained on Huawei Ascend 910B NPU released. May 22, 2026. https://new.nat6.net/en/news/cmph1mhwi0kj7sljwklnw8x2l/