Citation-impact pairwise judgment
According to the model card, the model compares two papers' titles, abstracts, and publication dates to predict higher citation impact.
Open Source Model Profile · OpenMOSS-Team
SciJudge-4B
SciJudge-4B is a 4.02B-parameter Qwen3 text-generation fine-tune from OpenMOSS-Team. According to the model card, it predicts which of two scientific papers has higher citation impact from titles, abstracts, and dates.
- Publisher
- OpenMOSS-Team
- Task
- text-generation
- Model type
- qwen3
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
SciJudge-4B is published by OpenMOSS-Team as a Qwen3 text-generation model. The captured configuration identifies Qwen3ForCausalLM with model type qwen3, and Safetensors metadata reports 4,022,468,096 parameters. According to the model card, it is a Qwen3-4B-Instruct-2507 fine-tune for scientific paper evaluation and part of the AI Can Learn Scientific Taste work with SciJudgeBench.
Recorded capabilities
Documented Qwen3 base
According to the model card, the base model is Qwen/Qwen3-4B-Instruct-2507; hub tags repeat that finetune linkage.
GRPO with DAPO training record
According to the model card, training used GRPO with DAPO loss, an external preference reward for citation-based pairwise judgment, bfloat16 precision, and KL coefficient 0.03.
Newer-release notice
According to the model card, a newer release is available at SciJudge-4B-2605 and is recommended for current experiments and comparisons.
Use cases in the source record
- Pairwise scientific-paper comparison that predicts higher citation impact from two titles, abstracts, and publication dates.
- SciJudgeBench-linked evaluation experiments described in the publisher's scientific-taste project.
Limitations and unknowns
- No extracted benchmark scores were included in this record despite the SciJudgeBench dataset reference.
- No context-window value, VRAM requirement, or quantization detail was extracted.
- Training information comes from the publisher model card and was not independently verified.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: OpenMOSS-Team/SciJudge-4B
Captured: Unknown. Processed: 2026-09-07T19:35:50.856938+00:00.
SciJudge-4B Update: A newer release is available at SciJudge-4B-2605 . We recommend using the newer release for current experiments and comparisons. SciJudge-4B is a Qwen3-4B-Instruct-2507 model fine-tuned for scientific paper evaluation. Given two papers' titles, abstracts, and publication dates, it predicts which paper has higher citation impact. This model is part of AI Can Learn Scientific Taste . The benchmark dataset is SciJudgeBench . Resources: Project page and GitHub repository . Usage import torch from transformers import AutoModelForCausalLM, AutoTokenizer model_name = "OpenMOSS-Team/SciJudge-4B" tokenizer = AutoTokenizer…
F001F002F003F004F005F006F007F008F009F010F011F012F013F015