Archives
- 18 Nov Five kernels against SGLang: my Muse Spark Open Kernel Challenge
- 03 Nov Third vLLM PR merged: embedding LoRA ignored embed_scale
- 28 Oct First SGLang PR merged: MegaMoE double-divided the DP prefill budget
- 22 Sep Second vLLM PR merged: NaN in the Triton softcap
- 11 Sep First vLLM PR merged: rsLoRA scaling in MoE expert packing
- 15 Aug vLLM vs SGLang — Benchmarking LLM Serving Engines on an RTX 4090
- 12 Aug PitchTwin