Downloads · 30 days
3.4K
32% of all-time downloads
Gammaception/Qwen3.5-27B-Thireus-16gb-optimized-GGUF
Qwen3.5-27B-Thireus-16gb-optimized-GGUF is a machine learning model from Gammaception. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Optimized llama.cpp GGUF quants of Qwen 3.5 27b made for 16gb VRAM cards.
Downloads · 30 days
3.4K
32% of all-time downloads
All-time downloads
10.8K
Public
Repo size
26.7 GB
Likes
9
Public
Click a slice to open those files.
.gguf26.1 GB · 98%
From the Hugging Face model README
Optimized llama.cpp GGUF quants of Qwen 3.5 27b made for 16gb VRAM cards.
Made using GGUF-Tool-suite! by Thireus, tuned by me :) (Gammaception)
Best 16 gb config for GC IQ3_M: (headless 64k ctx kv-cache@(q8_0,q8_0), -ub 256, q8_0 mmproj). Following official sampling parameters recommended, along with reasoning budget + message
EDIT 24/03/26: Updated GGUF to fix coding quality, new charts
EDIT 03/04/26: New variants with better quality ratios, update charts
| unsloth ud-iq3_xxs | bartowski iq3_xs | GC-Thireus iq3_xs | mungert iq3_xs | bartowski iq3_m | GC-Thireus iq3_m | unsloth ud-q3_k_xl | bartowski iq4_xs | |
|---|---|---|---|---|---|---|---|---|
| Size (GiB) | 10.716 | 11.357 | 11.800 | 12.078 | 12.240 | 12.465 | 13.447 | 14.130 |
| PPL mean | 7.349445 | 7.275691 | 7.003592 | 7.146062 | 7.166074 | 6.886104 | 7.095870 | 6.932149 |
| KLD mean | 0.080486 | 0.066744 | 0.044320 | 0.053966 | 0.057891 | 0.032533 | 0.034783 | 0.019950 |
| Same top p | 88.546 | 89.837 | 91.560 | 90.625 | 90.782 | 92.926 | 92.656 | 94.610 |
kl logits from unsloth ud-q6_k_xl (best i can fit), wikitext-2 512ctx bf16 kv
