The news of September 20thChina TelecomA new generation of stars was launched on September 17thLarge Model Xing4.0-29B-A4B, the total size of the model 29B, activating only 4B, original support 256K context, extended to 512KThe country ' s first large-scale model of 10 billion parameters to optimize the depth of complex engineering tasks, based on the country ' s economic capacity and the national framework.
Officially, from training chips to reasoning deploymentIt's all Chinese - Training, adaptation of the national production framework, self-study structuresOpen SourceEco, one way through。
According to the SuperCLUE assessment, in terms of intelligence capabilities, Xing4.0-29B-A4B scored 93.52, ranked third, and the gap with both head Qwen models was less than one point。
It is described that Xing4.0-29B-A4B can be reduced to 15GB (approximately 75% compared to FP16), RTX 3090, RTX 4090, a consumption scale of 24G, by quantifying 4-bit, to enable local context missions。
Xing4.0-29B-A4B Open source routes for the continuation of the Big Star Model provide full compatibility with mainstream open source ecology: support for fine-tuning structures such as LLAMA-Factory, MindFormers; fit for mainstream reasoning frameworks such as SGLang, vLM, KTransformers; and the Agent Framework provides orientation adaptations to OpenCode, Claude Code, OpenClaw, Hermes and deep alignment of framework formats。
Xing4.0-29B-A4B, an open-source community, 1AI with relevant links:
- HuggingFace: https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B
- GitHub: https://github.com/XingChen-AGI/Xing4.0-29B-A4B
- Magic: https://modelsscope.cn/models/XingChen-AGI/Xing4.0-29B-A4B
- Gitee: https://gitee.com/xingchen-agi/xing4.0-29b-a4b
- Magic: https://moders.cn/models/XingChen-AGI/Xing4.0-29B-A4B