Meta vision, speech models land on Ascend
Alibaba's ModelScope has released compatibility updates that port Meta's open-source vision and speech models to Huawei's Ascend NPUs. This release enables developers in China's AI ecosystem to run and fine-tune Meta's multimodal architectures natively on domestic hardware.
Porting Meta's leading open-weights models to Ascend NPUs is a strategic imperative for China's AI ecosystem as it rapidly decouples from NVIDIA hardware.
- –NVIDIA Decoupling: Accelerates the transition of Chinese AI developers to domestic hardware like Ascend 910B chips amid strict US export controls.
- –Turnkey Multimodal Capabilities: Puts out-of-the-box support for complex vision and speech tasks directly into developer hands without manual adaptation hurdles.
- –Ecosystem Integration: Leverages frameworks like ModelScope SWIFT to bridge the CANN-to-CUDA gap, providing a smooth path for PyTorch-dependent teams.
- –ModelScope's Hub Growth: Solidifies ModelScope's position as China's de facto Hugging Face equivalent by driving local hardware adaptation of global models.
DISCOVERED
2h ago
2026-06-25
PUBLISHED
2h ago
2026-06-25
RELEVANCE
AUTHOR
AiChinaNews