Skip to content
Wei-Cheng (Wayne) Chiu

Wei-Cheng Chiu

邱偉誠

CV

Taipei, Taiwan · UTC+8

making every FLOP count

Blog

I use these notes to answer the questions that survive the launch post: where the latency comes from, which claim the evidence actually supports, and what breaks when an architecture meets a real runtime. Equations, measurements, and source notes stay next to the claim they qualify.

Topics:
Inkling 官方發布視覺Thinking Machines Inkling 975B:5:1 混合注意力如何重新分配算力 5:1 滑動視窗、相對位置編碼、短卷積與 Muon/Adam:查核官方資料後拆解 Thinky 的第一個開放權重模型。 Kimi K3 官方發布視覺Moonshot AI Kimi K3 2.8T:KDA、AttnRes 與 896 選 16 的 MoE KDA、Attention Residuals 與 896 選 16 的 Stable LatentMoE;並釐清 API 已上線、權重尚待釋出的差別。 NVIDIA TensorRT-LLM 與 Triton 部署架構NVIDIA TensorRT-LLM vs vLLM on 4×H100|Concurrency crossover Matched-work benchmark:TensorRT-LLM 在低/中 concurrency 領先,vLLM 在高 concurrency 領先。 NVIDIA NCCL 官方主視覺NVIDIA Where tensor-parallel inference hits the NVLink wall|TP inference 的 NVLink 牆 366 GB/s 與 23 μs 固定成本:分開量測 NVLS、CUDA Graphs 和 symmetric memory 改變了什麼。 Microsoft 單租戶 RAG 架構圖Microsoft Learn The 0% RAG result failed the harder test|0% 結果沒有通過更難的測試 簡單資料集是 0%,adversarial near-miss 是 48%;比較九種 gates 後,grounded verifier 補抓到 generative judges 漏掉的錯誤。 Google Research 聯邦學習與差分隱私示意圖Google Research FedAvg Under Non-IID Data|DP-SGD 與 Secure Aggregation 保護什麼 從 client drift 到 privacy bound:FedAvg、FedProx、SCAFFOLD、DP-SGD 與 secure aggregation 的實測限制。 NVIDIA WMMA warp tile 結構圖NVIDIA From Naive GEMM to WMMA|每個 CUDA kernel 卡在哪裡 比較 naive、shared-memory tiling 與 WMMA,再用相同 shape 的 cuBLAS 當 production baseline。 DeepMind meta-reinforcement learning 實驗主視覺Google DeepMind Meta-RL:讓 policy 在 episode 內學會更新自己 把 task inference 放進 gradient、recurrent state 或 latent context;以及怎麼測出 adaptation 而不是 task memorization。 NVIDIA CUDA accelerated computing 官方視覺NVIDIA CUDA Programming 入門 從 execution model、memory hierarchy 與 synchronization 判斷 CUDA kernel 的實際瓶頸。 Kali Linux 2026.2 官方桌面視覺Kali Linux 滲透測試不是掃描器:從授權邊界到可複測證據 先寫 Rules of Engagement,再談工具;用最小化驗證、證據鏈與 acceptance test 把 finding 交給修復者。
Image sources Artwork and diagrams are credited to their original publishers: [Thinking Machines](https://thinkingmachines.ai/news/introducing-inkling/), [Moonshot AI](https://platform.kimi.ai/), [NVIDIA Developer](https://developer.nvidia.com/blog/), [Microsoft Learn](https://learn.microsoft.com/en-us/azure/architecture/ai-ml/guide/secure-multitenant-rag), [Google Research](https://research.google/blog/federated-learning-with-formal-differential-privacy-guarantees/), [Google DeepMind](https://deepmind.google/blog/prefrontal-cortex-as-a-meta-reinforcement-learning-system/), and [Kali Linux](https://www.kali.org/blog/kali-linux-2026-2-release/).