MagicQuant GGUF Hybrids - Qwen3 4B Thinking 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
mxfp4_moe-O-Q6K-EQKUD-Q8_0	3.90	369.09	0.0989%
mxfp4_moe-Q-Q5K-EKOUD-Q6K	3.03	394.06	0.1278%
iq4_nl-EQKOUD-Q6K	3.08	413.99	0.1740%
mxfp4_moe-QK-IQ4NL-O-MXFP4-EUD-Q6K	2.84	430.23	0.3832%
Q5_K	2.69	375.72	0.5973%
Q4_K_M	2.33	366.54	1.6668%
mxfp4_moe-QKU-IQ4NL-O-MXFP4-D-Q5K-E-Q6K	2.30	412.13	2.2740%
IQ4_NL	2.23	450.75	2.4657%
mxfp4_moe-EQOU-IQ4NL-KD-Q6K	2.37	472.25	2.5049%

Table - PPL Columns

model_name	gen	gen_er	code	code_er	math	math_er
mxfp4_moe-O-Q6K-EQKUD-Q8_0	10.0081	0.2450	1.5936	0.0128	6.9001	0.1413
mxfp4_moe-Q-Q5K-EKOUD-Q6K	9.9957	0.2441	1.5922	0.0127	6.9036	0.1412
iq4_nl-EQKOUD-Q6K	9.9687	0.2431	1.5927	0.0127	6.8924	0.1409
mxfp4_moe-QK-IQ4NL-O-MXFP4-EUD-Q6K	10.0858	0.2460	1.5949	0.0126	6.9032	0.1403
Q5_K	10.0993	0.2473	1.5978	0.0128	6.9256	0.1413
Q4_K_M	10.3239	0.2536	1.6093	0.0129	6.9423	0.1412
mxfp4_moe-QKU-IQ4NL-O-MXFP4-D-Q5K-E-Q6K	10.4164	0.2569	1.6143	0.0130	6.9825	0.1423
IQ4_NL	10.3718	0.2548	1.6125	0.0129	7.0606	0.1452
mxfp4_moe-EQOU-IQ4NL-KD-Q6K	10.3780	0.2547	1.6178	0.0132	7.0415	0.1443

Table - Precision Loss Columns

model_name	loss_general	loss_code	loss_math
mxfp4_moe-O-Q6K-EQKUD-Q8_0	0.0250	0.1194	0.1524
mxfp4_moe-Q-Q5K-EKOUD-Q6K	0.1488	0.0314	0.2032
iq4_nl-EQKOUD-Q6K	0.4186	0.0628	0.0406
mxfp4_moe-QK-IQ4NL-O-MXFP4-EUD-Q6K	0.7512	0.2010	0.1974
Q5_K	0.8861	0.3832	0.5225
Q4_K_M	3.1297	1.1057	0.7649
mxfp4_moe-QKU-IQ4NL-O-MXFP4-D-Q5K-E-Q6K	4.0537	1.4199	1.3484
IQ4_NL	3.6082	1.3068	2.4820
mxfp4_moe-EQOU-IQ4NL-KD-Q6K	3.6701	1.6398	2.2048

Baseline Models (Reference)

Table - File Size + TPS + Avg Precision Loss

model_name	file_size_gb	bench_tps	avg_prec_loss
BF16	7.50	249.86	0.0000%
Q8_0	3.99	360.78	0.1028%
Q6_K	3.08	404.72	0.1740%
Q5_K	2.69	375.72	0.5973%
Q4_K_M	2.33	366.54	1.6668%
IQ4_NL	2.23	450.75	2.4657%
MXFP4_MOE	2.00	466.66	7.9498%

Table - PPL Columns

model_name	gen	gen_er	code	code_er	math	math_er
BF16	10.0106	0.2451	1.5917	0.0127	6.8896	0.1410
Q8_0	10.0174	0.2454	1.5931	0.0128	6.9001	0.1413
Q6_K	9.9687	0.2431	1.5927	0.0127	6.8924	0.1409
Q5_K	10.0993	0.2473	1.5978	0.0128	6.9256	0.1413
Q4_K_M	10.3239	0.2536	1.6093	0.0129	6.9423	0.1412
IQ4_NL	10.3718	0.2548	1.6125	0.0129	7.0606	0.1452
MXFP4_MOE	10.9465	0.2659	1.6645	0.0138	7.5735	0.1563

Table - Precision Loss Columns

model_name	loss_general	loss_code	loss_math
BF16	0.0000	0.0000	0.0000
Q8_0	0.0679	0.0880	0.1524
Q6_K	0.4186	0.0628	0.0406
Q5_K	0.8861	0.3832	0.5225
Q4_K_M	3.1297	1.1057	0.7649
IQ4_NL	3.6082	1.3068	2.4820
MXFP4_MOE	9.3491	4.5737	9.9266

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 :)