Artificial Intelligence thread
- Thread starter 9dashline
- Start date
The first time for domestic GPUs! Moore Threads and Wuwen Core Qyon have achieved large-scale model training on domestic GPUs for the first time!
Moore Threads and Wuwen Xinqiong announced that the two parties have officially completed the 3B-scale large-scale model training based on the domestic full-featured GPU Qianka cluster. The model is called "MT-infini-3B" and has completed efficient and stable training on Moore Thread's Kua'e (KUAE) Qianka Intelligent Computing Cluster and Wuwen Xinqiong AIStudio PaaS platform. This training fully verified the reliability of the Kua'e Qianka Intelligent Computing Cluster in large-scale model training scenarios, and also took the lead in opening up a new paradigm of in-depth cooperation between domestic large language models and domestic GPU Qianka Intelligent Computing Clusters in the industry.
The total training time of the MT-infini-3B model was 13.2 days. After precision debugging, the training was stable and uninterrupted throughout the whole process. The stability of cluster training reached 100%, and the expansion efficiency of training exceeded 90% compared with single-machine training. At present, the performance of the trained MT-infini-3B ranks among the top models of the same scale. Compared with other models trained on international mainstream hardware, it has achieved leading performance on three test sets including C-Eval, MMLU, and CMMLU.
Huawei launched a series of Galaxy/Xinghe AI products for North African market from Marrakesh, Morocco today
in the future, please change x to twitter in there, so that it will show up correctlly