Chunk 65.4
From Diagnosis to Deployment: Three Threads of Inference Engineering for Kimi K2.6
Message Articles
- The 32 t/s Puzzle: Diagnosing a Throughput Regression in Speculative Decoding
- The Anatomy of a Performance Regression: Diagnosing DDTree Speculative Decoding at Scale
- The Art of Diagnostic Isolation: How a Controlled Experiment Unraveled a Speculative Decoding Mystery
- The Syntax Error That Almost Solved It: Diagnosing Speculative Decode Throughput at Scale
- The Decisive Experiment: Diagnosing DDTree Speculative Decoding Throughput at Scale
- The Anatomy of a Performance Diagnosis: Unraveling the 32 t/s Mystery in Speculative Decoding
- The Quiet Capstone: How a One-Line Git Verification Crystallized a Complex Performance Diagnosis
- The Verdict on 32 Tok/s: How Rigorous Diagnosis Separated System Behavior from Perceived Regression
- The Six-Word Pivot: "what's the max context? Set to 200k?"
- The 200k Context Question: Diagnosing Feasibility Before Action
- Pushing the Context Window: Diagnosing Memory Feasibility for 200k-Token Inference on Blackwell GPUs
- The 200k Context Decision: Extending Kimi K2.6's Reach on Blackwell GPUs
- The Empty Message: Silence as Signal in AI-Assisted Coding Sessions