released Xing 4 large language models
The story so far
Huawei has released the Xing 4 large language model, extending the capabilities of the TeleChat series on Ascend chips and the MindSpore framework. Xing 4, through the collaborative progress of architecture and training infrastructure, supports pre-training 285B parameter models on 8192 Ascend 910C GPUs and introduces the TeleGym unified agent execution platform to optimize strategies. Its pre-training infrastructure is based on MindSpore HyperParallel, combining topological-aware hybrid sharding, efficient distributed optimizer execution, and coordinated management of long sequence workloads and activation memory. The post-training phase introduces TeleGym, connecting execution, evaluation, and trajectory construction for strategy optimization, using asynchronous execution, bounded oversampling, and partial release to adapt to the duration of heterogeneous tasks while maintaining trajectory continuity between strategy updates.
AI-written from the reports · Updated 16 hours ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hugging Face Daily PapersTeleChat to Xing4: Advancing Frontier Agentic Models on Domestic Chips and Frameworks
Xing4 is a series of LLMs developed on Huawei Ascend chips and the MindSpore framework, expanding the capabilities of the TeleChat series through the synergistic progress of architecture and training infrastructure, supporting the pre-training of 285B parameter models on 8192 blocks of Ascend 910C GPUs, and introducing the TeleGym unified agent execution platform to optimize strategies.
Heat over time
- Now
- 6
- Peak
- 818 Nov, 14:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.