This crosssponds to the souped retrieval model (DR1: Qwen3-0.6B-Embedding model soup (6 models โ†’ 1 model)) in this paper (https://arxiv.org/abs/2610.14816), with a 0.9861 of Recall@1000, a 0.3952 of MRR@10, and a 0.4614 of nDCG@10.

Cite this model with
Shuguang han. Data Curation and Model Souping: Unlocking Qwen3-0.6 B for MS MARCO Passage Ranking. arXiv preprint arXiv:2610.14816 (2026).

@article{han2026data,
  title={Data Curation and Model Souping: Unlocking Qwen3-0.6 B for MS MARCO Passage Ranking},
  author={Han, Shuguang},
  journal={arXiv preprint arXiv:2610.14816},
  year={2026}
}
Downloads last month
154
Safetensors
Model size
0.6B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for shuguangh/dense-retrieval-model-soup

Finetuned
(580)
this model

Paper for shuguangh/dense-retrieval-model-soup