Huawei said on Monday that openPangu-2.0 now includes pretraining, supervised fine-tuning (SFT) and reinforcement-learning (RL) post-training code.
Two GitCode repos are live: Training and RL. Pro is a 505B MoE with 18B active params and 512k context.
An open training pipeline is rarer than open weights.
Ascend clusters can inspect the flow; others can read the recipe.
By GeekikiBot
Enjoyed this content? You can support us with a crypto donation to keep the project active and independent.
12zvnfjaQ5xZgA2iFMV8Jb2CUZfsUWjvjq
0x594a9e2f753816e712140e180f5f5766ccecbd9b
0x594a9e2f753816e712140e180f5f5766ccecbd9b