This draft model is OP on 16 GB cards for Qwen 3.8 27b

A user reported that a draft model, specifically HermiHg/Qwen3.8-27B-DFlash2-Q2_K_S-MIX-GGUF, performs exceptionally well on 16 GB cards. When paired with ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF (IQ3_XXS with 128k context), it achieved an average token generation speed of about 60 tokens per second on an RX 9070 XT graphics card. Further testing by another individual on a different card is also mentioned.

This report uniquely details specific performance metrics for the HermiHg/Qwen3.8-27B-DFlash2-Q2_K_S-MIX-GGUF draft model on 16 GB cards, unlike general discussions of model efficiency.