Extremely slow DSpark draft model performance (1-2 t/s) with DeepSeek-V4-Flash on llama-server compared to MTP?

Hey everyone, I could use some advice on setting up speculative decoding correctly with llama-server. My Hardware: GPUs: RTX 4090 + RTX 6000 Pro (120GB total VRAM) RAM: 32GB I am currently testing the DeepSeek-V4-Flash-0731 unsloth model using the 137GB Q4 UD-IQ4-NL quant. When running the MTP draft