No official torrent yet — requesting it tells the flock what to host next. You cannot seed a catalog-only page.
Cataloged from ISTA-DASLab/Qwen3-30B-A3B-P48NVFP4-MoESQ
All models by ISTA-DASLab →A W4A4 + paired-4:8 sparse compressed checkpoint of
Qwen/Qwen3-30B-A3B, produced with MoESQ.
The MoE expert weights are NVFP4 with paired-4:8 structured sparsity. They are stored
sparse: only the kept values plus a small mask are on disk, not a dense NVFP4 tensor with
zeros in place. The target is NVIDIA Blackwell (SM100 and SM120) sparse tensor cores.
paired48_nvfp4 MoE backendLinks
This checkpoint does not load in upstream vLLM. The MoESQ repository installs a patched
vLLM v0.30.0 (the paired48_nvfp4 MoE backend and sparse-storage loader) together with the
kernels. It needs an NVIDIA Blackwell SM100 (B200, GB200) or SM120 (RTX 5090, RTX PRO 6000) GPU and a
CUDA toolkit >= 12.8; SM103 (B300) and SM121 (DGX Spark) are not supported.
vLLM selects the backend automatically.
The six tasks are ARC-Challenge (25-shot), HellaSwag (10-shot), MMLU (5-shot), TruthfulQA-MC2 (0-shot), Winogrande (5-shot) and GSM8K (5-shot, greedy). Scores use lm-evaluation-harness with the full test sets, mean ± sd over few-shot seeds 1234, 0 and 1, served with the MoESQ vLLM integration on 1× B200. Recovery is relative to dense.
Per task, this model: ARC-C 60.78 · GSM8K 82.36 · HellaSwag 64.66 · MMLU 71.13 · TruthfulQA-MC2 51.73 · Winogrande 67.80.
Gate and up projections carry separate NVFP4 global scales. The MoESQ vLLM patch folds their ratio into the block scales at load time. A vLLM build without that fold applies the gate scale to both projections; this checkpoint's weights scored 65.20 ± 0.08 on such a build under the same protocol.
The scale group is 32 because the Blackwell sparse NVFP4 MMA (SM100 and SM120) requires one scale per 32 dense K elements, which is 16 surviving elements after the 4:8 prune.
paired48_sparse, pair-bitmask v1)Packed NVFP4 stores two adjacent K elements per byte, so one pair is one byte. A paired-4:8 weight therefore has exactly 2 nonzero bytes in every 4-byte chunk. Each routed-expert linear stores:
quantization_config carries "paired48_sparse": {"layout": "pair-bitmask", "version": 1}.
The encoding is lossless: converting to and from dense NVFP4 is exact, and every tensor in
this upload was checked with a round trip. At load time vLLM rebuilds each layer's dense
packed weight and compresses it into the kernel's own layout.
The full MoESQ config for this run is in moe_sq_config.yaml [blocked].
gate_weight_exponent = 2).For questions, open a discussion here or contact kwanhee.lee@postech.ac.kr.
git clone --recurse-submodules https://github.com/IST-DASLab/MoESQ.git && cd MoESQbash integrations/vllm/install.sh && source .venv-vllm/bin/activate
vllm serve ISTA-DASLab/Qwen3-30B-A3B-P48NVFP4-MoESQ # 1x B200# multi-GPU: add --tensor-parallel-size N --enable-expert-parallel@misc{lee2026hardwarenativejointsparsequantizationtrillionscale, title={Hardware-Native Joint Sparse-Quantization for Trillion-Scale Mixture-of-Experts}, author={Kwanhee Lee and Namhoon Lee and Dan Alistarh}, year={2026}, eprint={2610.02241}, archivePrefix={arXiv}, primaryClass={cs.AR}, doi={10.48550/arXiv.2610.02241}, url={https://arxiv.org/abs/2610.02241},}