LLM SERVING

A fair comparison of vLLM, TGI, SGLang, TensorRT-LLM, llama.cpp, and LMDeploy for self-hosted LLM inference - batching, quantization, hardware, and ops.