MagicQuant GGUF Hybrids - Qwen3 30B A3B Instruct 2507

MagicQuant is an automated quantization, benchmarking, and evolutionary hybrid-GGUF search system for LLMs.

Each release includes models optimized to outperform standard baseline quants (Q8, Q6, Q5, Q4). If a baseline GGUF exists in this repo, the evolutionary engine couldn’t beat it. If a baseline is missing, it’s because a hybrid configuration outperformed it so completely that including the baseline would've been pointless.

These hybrid GGUFs are built to be as small, fast, and low-drift as possible while preserving model capability.

To dive deeper into how MagicQuant works, see the main repo: MagicQuant on GitHub (by MagicCodingMan)

Notes:

The HuggingFace hardware compatibility where it shows the bits is usually wrong. It doesn't understand hybrid mixes, so don't trust it.
Naming scheme can be found on the MagicQuant Wiki.
(tips) Less precision loss means less brain damage. More TPS means faster! Smaller is always better right?

Precision Loss Guide

0–0.1% → God-tier, scientifically exact
0.1–1% → True near-lossless, agent-ready
1–3% → Minimal loss, great for personal use
3–5% → Borderline, but still functional
5%+ → Toys, not tools, outside MagicQuant’s scope

Learn more about precision loss here.

Table - File Size + TPS + Avg Precision Loss

model_name	file_size_gb	bench_tps	avg_prec_loss
iq4_nl-EHQKOUD-Q8_0	30.25	99.68	0.0771%
Q5_K	20.23	117.37	0.2007%
mxfp4_moe-H-B16-EUR-IQ4NL-KO-Q5K-QD-Q6K	18.93	110.54	0.3929%
IQ4_NL	16.26	138.69	0.4198%
iq4_nl-EHQKOUD-IQ4NL	16.04	149.76	2.6323%

Table - PPL Columns

model_name	gen	gen_er	code	code_er	math	math_er
iq4_nl-EHQKOUD-Q8_0	6.2536	0.1277	1.2991	0.0072	5.7045	0.1063
Q5_K	6.2777	0.1283	1.3006	0.0073	5.7037	0.1062
mxfp4_moe-H-B16-EUR-IQ4NL-KO-Q5K-QD-Q6K	6.2854	0.1284	1.3036	0.0072	5.7274	0.1068
IQ4_NL	6.2669	0.1274	1.3111	0.0073	5.7159	0.1061
iq4_nl-EHQKOUD-IQ4NL	6.4836	0.1337	1.3170	0.0075	5.8712	0.1099

Table - Precision Loss Columns

model_name	loss_general	loss_code	loss_math
iq4_nl-EHQKOUD-Q8_0	0.0719	0.0770	0.0823
Q5_K	0.3132	0.1926	0.0963
mxfp4_moe-H-B16-EUR-IQ4NL-KO-Q5K-QD-Q6K	0.4362	0.4237	0.3188
IQ4_NL	0.1406	1.0015	0.1174
iq4_nl-EHQKOUD-IQ4NL	3.6033	1.4560	2.8375

Baseline Models (Reference)

Table - File Size + TPS + Avg Precision Loss

model_name	file_size_gb	bench_tps	avg_prec_loss
BF16	56.90	44.48	0.0000%
Q8_0	30.25	95.03	0.0771%
Q5_K	20.23	117.37	0.2007%
Q6_K	23.37	108.10	0.3089%
IQ4_NL	16.26	138.69	0.4198%
Q4_K_M	17.28	132.46	1.4766%
MXFP4_MOE	15.15	138.34	9.0818%

Table - PPL Columns

model_name	gen	gen_er	code	code_er	math	math_er
BF16	6.2581	0.1279	1.2981	0.0072	5.7092	0.1064
Q8_0	6.2536	0.1277	1.2991	0.0072	5.7045	0.1063
Q5_K	6.2777	0.1283	1.3006	0.0073	5.7037	0.1062
Q6_K	6.2881	0.1290	1.3002	0.0072	5.7255	0.1072
IQ4_NL	6.2669	0.1274	1.3111	0.0073	5.7159	0.1061
Q4_K_M	6.4032	0.1315	1.3145	0.0074	5.7576	0.1073
MXFP4_MOE	7.0161	0.1472	1.3631	0.0083	6.2873	0.1213

Table - Precision Loss Columns

model_name	loss_general	loss_code	loss_math
BF16	0.0000	0.0000	0.0000
Q8_0	0.0719	0.0770	0.0823
Q5_K	0.3132	0.1926	0.0963
Q6_K	0.4794	0.1618	0.2855
IQ4_NL	0.1406	1.0015	0.1174
Q4_K_M	2.3186	1.2634	0.8478
MXFP4_MOE	12.1123	5.0073	10.1258

Support

I’m a solo developer working full time for myself to achieve my dream, pouring nights and weekends into open protocols and tools that I hope make the world a little better. If you chip in, you're helping me keep the lights on while I keep shipping.

Click here to see ways to support - BTC, Paypal, GitHub sponsors.

Or, just drop a like on the repo :)