Linux kernel and systems fast path for LLM inference: eBPF tracing, runtime hints, cgroups, NUMA/GPU locality, KV-cache memory policies, TTFT boost, and experimental kernel primitives.
performance kernel gpu scheduler linux-kernel systems cgroups numa ebpf observability inference-optimization kv-cache ttft llama-cpp vllm llm-inference tensorrt-llm sglang tokens-per-second kernel-performance
-
Updated
Jul 6, 2026