Qwen 3.6 27B and 35B MTP vs Standard on 16GB GPU

I tested Speculative decoding (Multi-Token Prediction, MTP) performance in Qwen 3.6 27B and 35B on an RTX 4080 with 16 GB VRAM. For a broader view of token speeds and VRAM trade-offs across more model...