Downloads · 30 days
23
11% of all-time downloads
neopolita/Qwen3.6-35B-A3B-8bit-Prune-30-Uniform
Qwen3.6-35B-A3B-8bit-Prune-30-Uniform is a machine learning model from neopolita. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as apache-2.0.
A 30%-expert-pruned and router-healed MLX 8-bit derivative of mlx-community/Qwen3.6-35B-A3B-8bit.
Downloads · 30 days
23
11% of all-time downloads
All-time downloads
211
Public
Parameters
25B
26.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors26.5 GB · 100%
How the weights are stored.
U3225B · 100%
From the Hugging Face model README
A 30%-expert-pruned and router-healed MLX 8-bit derivative of mlx-community/Qwen3.6-35B-A3B-8bit.
30% of routed experts (256 → 179/layer, uniform) were removed by English-cold frequency selection (drop the experts English routes to least, per layer), then routers were recalibrated with Router-KD — distilling the dense teacher's next-token distribution into the surviving routers only (rank-64 LoRA on mlp.gate, ~5.7M params, wikitext calibration), then fused.
| Routed experts | 256 → 179/layer (−30%), uniform |
| Active | top-8 + 1 shared |
| Healing | Router-KD (router-only) |
| Format | MLX, 8-bit |
| metric | base | this model |
|---|---|---|
| wikitext PPL | 7.98 | 8.54 |
| winogrande | 0.70 | 0.72 |
| hellaswag (acc_norm) | 0.555 | 0.575 |
| piqa (acc_norm) | 0.84 | 0.81 |
| gsm8k (strict) | 0.99 | 0.98 |
| ifeval (prompt-strict) | 0.86 | 0.86 |
| ifeval (inst-loose) | 0.933 | 0.933 |
At 30% the prune is near-free on tasks, and Router-KD fully recovers instruction-following (ifeval).
from mlx_lm import load, generate
model, tok = load("neopolita/Qwen3.6-35B-A3B-8bit-Prune-30-Uniform")
print(generate(model, tok, prompt="Explain MoE pruning in one sentence.", max_tokens=128))
Research artifact. Selection by per-layer English routing frequency (not REAP); small-sample benchmarks (100–200). License/usage inherit from the base model.