Qwen3-30B-A3B-Dense-3B-DO-ACP-K8
Research artifact, not a general-purpose model. These weights are released so that the experiments in the paper Pruning and Distilling Mixture-of-Experts into Dense Language Models can be reproduced. Each student is distilled on a fixed, small token budget (~4B tokens for Qwen3, 0.3B tokens for DeepSeek-V2-Lite and GPT-OSS) purely so that expert scoring and grouping methods can be compared under an equal budget. Absolute quality is therefore far below the teacher and below pretrained models of the same size, and no instruction tuning or alignment was applied. Please do not use this as an off-the-shelf assistant.
What this is
A dense student obtained by pruning and distilling a Mixture-of-Experts teacher, from the paper Pruning and Distilling Mixture-of-Experts into Dense Language Models. Code: https://github.com/krafton-ai/moe-to-dense
This is the main student of the paper and the best configuration found in the full scoring x grouping sweep.
Results
Full comparison group for this architecture, extended training after ~4B tokens (paper Table 6). This model's row is in bold, and rows whose weights are also released link to them.
| Configuration | Wino | Hella | ARC-E | ARC-C | MMLU | Avg |
|---|---|---|---|---|---|---|
| DO-ACP, K=8 | 63.1 | 60.3 | 75.6 | 45.4 | 46.1 | 58.10 |
| SF, K=16 | 61.2 | 56.3 | 74.0 | 43.1 | 32.7 | 53.46 |
| D2D pruning (Qwen3-32B to 3.4B) | 60.5 | 57.5 | 73.1 | 41.5 | 26.6 | 51.84 |
| Random FFN + teacher attn | 54.4 | 45.4 | 66.0 | 34.2 | 27.1 | 45.44 |
| Qwen3-1.7B (pretrained reference) | 66.1 | 67.1 | 81.9 | 55.5 | 62.6 | 66.63 |
| Qwen3-4B (pretrained reference) | 72.0 | 75.8 | 86.2 | 64.6 | 73.1 | 74.34 |
Downstream accuracy is Winogrande 5-shot, HellaSwag 10-shot, ARC-Easy 25-shot, ARC-Challenge 25-shot and MMLU 5-shot. Avg is the unweighted mean of the five benchmarks.
Configuration
| Field | Value |
|---|---|
| Teacher | Qwen/Qwen3-30B-A3B |
| Student parameters | 3.34B |
| Distillation data | FineWeb-Edu (sample-10BT), ~4B tokens |
| Expert scoring | DO-ACP |
| Experts kept (K) | 8 |
| Grouping | pure pruning (K equals the teacher's top-k, one expert per group) |
| Down-projection scaling | uniform |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("jhyuckkim/Qwen3-30B-A3B-Dense-3B-DO-ACP-K8", dtype="bfloat16")
tok = AutoTokenizer.from_pretrained("jhyuckkim/Qwen3-30B-A3B-Dense-3B-DO-ACP-K8")
The student is a standard Qwen3ForCausalLM, so it loads with stock
transformers and runs in vLLM without extra code.
Citation
@article{kim2026pruning,
title={Pruning and Distilling Mixture-of-Experts into Dense Language Models},
author={Kim, Junhyuck and Yun, Jihun and Kim, Haechan and Kim, Gyeongman and Bae, Joonghyun and Cho, Jaewoong},
journal={arXiv preprint arXiv:2605.28207},
year={2026}
}