Implemented and benchmarked LLM inference compression: int4/int8 quantization, GPTQ-like calibration, int8 KV cache, pruning, distillation, speculative decoding, torch.compile, and ONNX. Every number from a logged, self-audited run.