English summaries on zzszmyfhttps://zzszmyf.github.io/en/Recent content in English summaries on zzszmyfHugozh-CN© 2026 zzszmyfThu, 29 Oct 2026 00:00:00 +0800Why is FlashAttention faster and lighter on memory?https://zzszmyf.github.io/en/flashattention/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/flashattention/How does PagedAttention save KV-cache memory?https://zzszmyf.github.io/en/pagedattention/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/pagedattention/What is RadixAttention, and when does prefix cache help?https://zzszmyf.github.io/en/radixattention/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/radixattention/Why is EAGLE faster than Medusa?https://zzszmyf.github.io/en/eagle/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/eagle/AWQ or GPTQ for 4-bit weights?https://zzszmyf.github.io/en/awq-vs-gptq/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/awq-vs-gptq/FP8 or FP4 for a production LLM?https://zzszmyf.github.io/en/fp8-vs-fp4/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/fp8-vs-fp4/MQA, GQA, or MLA: what should you cache per token?https://zzszmyf.github.io/en/mqa-gqa-mla/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/mqa-gqa-mla/Why did a vLLM GPU stay at 100%?https://zzszmyf.github.io/en/vllm-max-tokens/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/vllm-max-tokens/vLLM, SGLang, or TensorRT-LLM?https://zzszmyf.github.io/en/vllm-vs-sglang-vs-tensorrt-llm/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/vllm-vs-sglang-vs-tensorrt-llm/Where do the quantization notes start?https://zzszmyf.github.io/en/quantization-series/Thu, 29 Oct 2026 00:00:00 +0800https://zzszmyf.github.io/en/quantization-series/