• Sep 18, 2026LLM Inference on a Budget: What Actually Moves the NeedleKV-cache, quantization, batching, and speculative decoding — the practitioner shortlist for serving LLMs fast without a GPU cluster.

All posts →