26.3.9
服务已重新上线.
更新chat后端到1.2.005:
- 支持基于tokenizer的开销计算
- 大量规范, 改良, 基建和性能优化工作
更换MF/MN/MV模型为Qwen3.5-35B-A3B于本地, 不再依赖上游提供商.
已知的硬件问题已修复, 但需要进一步观察.
此外, 一台设备(XP00)的散热系统有衰退迹象, 预计于一段时间后再次执行维护以维修更换. 具体另行通知.
26.3.9
Service backed online.
Updating chat backend to 1.2.005:
- Added support for tokenizer api implementation of consumation calculation
- Series of standardizations, improvements, constructions and optimizations
MFocus/MNerve/MVista model now migrated to local Qwen3.5-35B-A3B, no longer depend on 3rd provider.
Known hardware issue patched, but still need further observation.
Another potential maintenance required and will be scheduled ~1 month later, please wait for further announcements.