vllm 5
- Prompt Order Is a Cost Control: Measuring Prefix-Cache Reuse in Standard-Library Python
- Guess Ahead, Pay Once: Speculative Decoding You Can Verify
- Half the Cache, Twice the Context — measuring what KV quantization costs at long range
- How to Self-Host Open-Weight LLMs: A Practical Guide
- vLLM and High-Throughput LLM Serving: PagedAttention and Continuous Batching