88Z AI, 25 May — ModelBest, together with Tsinghua University and the OpenBMB open-source community, has released BitCPM-CANN, China's first 1.58-bit ternary large model trained end-to-end entirely on Huawei's Ascend domestic computing platform. Available in four sizes ranging from 0.5B to 8B, the model uses quantization-aware training to free up approximately six times the video memory during inference while retaining 90% to 97.2% of its capabilities. This release aims to provide critical infrastructure for edg
At a glance
- China's first 1.58-bit ternary large model trained end-to-end on Huawei Ascend is open-sourced
- Inference frees up ~6x video memory with 90%–97.2% capability retention
The update
ModelBest has partnered with Tsinghua University and the OpenBMB open-source community to officially open-source BitCPM-CANN. This is China's first 1.58-bit ternary large model trained end-to-end entirely on Huawei's Ascend domestic computing platform. The model series includes four different sizes: 0.5B, 1.5B, 4B, and 8B, catering to diverse deployment needs. Technically, BitCPM-CANN employs a quantization-aware training approach. This design allows the model to free up approximately six times the video memory during inference, significantly reducing hardware resource consumption. At the same time, the model retains 90% to 97.2% of its capabilities. This initiative not only improves operational efficiency but also provides critical infrastructure support for the practical implementation of edge-side AI and the enhancement of the domestic NPU ecosystem.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 面壁智能Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.