Anbeeld/Qwen3.6-27B-DFlash-GGUF

DFlash draft model for Qwen 3.6 27B, made by z-lab.

Tested with BeeLlama.cpp v0.2.0 — a llama.cpp fork with advanced DFlash support that enables using these draft models to their full potential.

• Target model: Qwen 3.6 27B Q5_K_S
• Setup: Windows 11, AMD Ryzen 7 5700X3D, 32 GB DDR4 RAM, RTX 3090 24 GB
• Config: same as in quick start docs, but with reasoning and adaptive DM disabled
• Baseline is llama.cpp b9275 CUDA 13.1 Windows prebuilt: 36.8 tok/s median

Prompt: Doubly-linked list (output: ~4K tok)

Write a complete Python 3 module implementing a doubly-linked list with the following methods: append, prepend, insert_at, remove_at, find, reverse, to_list, length, is_empty, iter. Include comprehensive docstrings, type hints, and pytest unit tests for every method. Return only the code, no commentary.

DFlash quant	Size	Median	Best	Speedup	Acceptance
IQ4_XS	891 MB	148.0 tok/s	160.5 tok/s	4.02x	47.6% / 87.7%
Q4_K_M	985 MB	145.6 tok/s	152.6 tok/s	3.96x	47.0% / 87.6%
Q5_K_M	1.17 GB	144.9 tok/s	157.2 tok/s	3.94x	46.8% / 87.6%
Q6_K	1.36 GB	139.2 tok/s	152.5 tok/s	3.79x	45.4% / 87.2%
Q8_0	1.76 GB	142.8 tok/s	155.5 tok/s	3.88x	46.9% / 87.6%
bf16	3.31 GB	132.5 tok/s	145.0 tok/s	3.60x	44.2% / 86.9%

Acceptance: accepted to proposed draft tokens / accepted draft tokens to final generated tokens

Between IQ4_XS, Q4_K_M and Q5_K_M the difference is smaller than noise from variance between passes, so using any of them should be fine. IQ4_XS takes up the least VRAM, but Q5_K_M might result in slightly higher acceptance in the long run.

Higher quants don't guarantee better performance: the model's job is to predict just a few tokens at the time, so loss of precision doesn't affect it as much. Meanwhile, larger size leads to slower drafting, reducing resulting tok/s, and also more VRAM consumption.

Keep in mind that results will likely be different for higher target model quants, which I can't test myself due to VRAM limitations.