Hugging Face's logo

Jundot
/
Qwen3.5-35B-A3B-oQ4e

MLX
Safetensors
qwen3_5_moe
oq
quantized
4-bit precision
Model card Files
xet
 Community
2

Thanks for this model

#1
by NapYang - opened
Discussion
NapYang

This is the best 4bit model. A quick test from your oMLX bench mark, humaneval pass 91%! (91/100).
rairenter

True!

here's my test

Intelligence Benchmark Comparison

          Mode    Sampled         Qwen3.5-35B-A3B-oQ4e

HUMANEVAL Full 164 90.8%
MBPP Sample 200/500 78.0%

--- Detail ---

Model: Qwen3.5-35B-A3B-oQ4e
Benchmark Accuracy Correct Total Time(s)

HUMANEVAL 90.8% 149 164 295.1
MBPP 78.0% 156 200 361.2

Sign up or log in to comment