Tag
Running Qwen3.6 27B on an RTX 5090, achieving 6.4k tokens per second after tuning MTP and cache settings, demonstrating optimization techniques for inference.