root 79a6a9faf7 docs: refine Unified Tensor Protocol with Adapter+Facade patterns and Zero-Copy advantages ...

- Added precise design pattern description (Adapter Pattern for codec, Facade Pattern for TensorBus interface)
- Dynamic grid-joining: register new subnets at runtime via TensorBus.register()
- Hot-plug: add/remove subnets without interrupting system
- Zero-Copy via kernel bypass on HarmonyOS 6.1
- Hardware affinity: NPU/GPU native tensor operations
- High data fidelity: lossless transfer of high-dim feature maps & attention weights
- Runtime training: support creating subnets on-demand while system runs

2026-06-01 09:23:05 +08:00

..

architecture-overview.md

docs: refine Unified Tensor Protocol with Adapter+Facade patterns and Zero-Copy advantages

2026-06-01 09:23:05 +08:00

llm-subnet.md

fix: correct architecture docs - tensor protocol, heuristic scheduling, LLM dual output, memory update mechanism

2026-06-01 09:20:14 +08:00

logic-trace-subnet.md

fix: correct architecture docs - tensor protocol, heuristic scheduling, LLM dual output, memory update mechanism

2026-06-01 09:20:14 +08:00

memory-subnet.md

fix: correct architecture docs - tensor protocol, heuristic scheduling, LLM dual output, memory update mechanism

2026-06-01 09:20:14 +08:00

routing-compression-subnet.md

fix: correct architecture docs - tensor protocol, heuristic scheduling, LLM dual output, memory update mechanism

2026-06-01 09:20:14 +08:00