BWhat changed
Co-designed a Unified-Bus supernode with independently scalable prefill, decode and caching pools, exploiting direct remote-memory access and expert-parallel serving for large MoE models.
Scientific scope
Co-designed a Unified-Bus supernode with independently scalable prefill, decode and caching pools, exploiting direct remote-memory access and expert-parallel serving for large MoE models. Prefill/decode separation, MoE and cache pooling have predecessors; contribution is the specified hardware/software integration. Network numbers can be per die rather than per dual-die NPU and unidirectional rather than bidirectional. Provider comparisons vary latency SLO, concurrency, precision, parallelism and hardware; no universal accuracy/throughput claim or independent replay. Original46-person Huawei/SiliconFlow affiliations are preserved.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Prefill/decode separation, MoE and cache pooling have predecessors; contribution is the specified hardware/software integration. Network numbers can be per die rather than per dual-die NPU and unidirectional rather than bidirectional. Provider comparisons vary latency SLO, concurrency, precision, parallelism and hardware; no universal accuracy/throughput claim or independent replay. Original46-person Huawei/SiliconFlow affiliations are preserved.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction