MTP has no impact on my Qwen3.6 MoE performance

In three linesUser reports MTP (Multi-Token Prediction) provides no performance improvement for Qwen3.6-35B GGUF on RTX 5060Ti: ~60 tok/s in both cases. Tests with unsloth flags (spec-type draft-mtp, spec-draft-n-max 2) but observes no speedup despite reducing ctx-size and quantization.

Summary generated by Claude — human-verified