← Back to all news
arXiv Computation and Language By Xuechen Li (University of Minnesota Twin Cities)

AgriJev: Efficient Agricultural Candidate Scoring with Multi-Scale Adapted Models

Read the original on arXiv Computation and Language →

AgriJev is a family of adapted Qwen3.5 models (0.8B, 2B, 4B, and 9B) designed for efficient agricultural candidate scoring. The models use a shared interface that conditions on agricultural text, a judgment question, and candidate definitions to return scores and relative probabilities in a single forward pass. Across 17 external test views, AgriJev-9B achieves 75.44% mean accuracy, surpassing both the original Qwen3.5 and matched agricultural SFT models, while also offering roughly half the median scoring latency of equivalent answer‑generation models on an A100 GPU.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.

Related stories

arXiv AI
Jun 9

AgroOmni: A Large-Scale Multi-view Agricultural Dataset for Cross-Scale Multimodal Reasoning

arXiv:2603. 14342v2 Announce Type: replace-cross Abstract: Modern agricultural data is sourced from diverse platforms and spans multiple spatial scales, ranging from ground-level close-up photography to Unmanned Aerial Vehicle (UAV) aerial observation and satellite remote sensing imagery.

By Jiarui Zhang, Junqi Hu, Zurong Mai, Yang Liu, Yuhang Chen, Shuohong Lou, Henglian Huang, Hong Cheng, Lingyuan Zhao, Jianxi Huang, Yutong Lu, Haohuan Fu, Juepeng Zheng
llmsnlpfine-tuningmultimodalbenchmarkssafety
More like this →
arXiv AI
Sep 18

QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training

QVAC Genesis III is a 191.43 B‑token synthetic STEM corpus covering 19 domains and multiple difficulty levels, created through a dual generation strategy that uses a weak edge‑scale student model to generate corrective explanations and contrastive reasoning. The authors evaluate the corpus with an LLM‑as‑a‑parser protocol and demonstrate that 1.7 B‑parameter models trained on QVAC Genesis III outperform those trained on Cosmopedia‑v2 and the Cosmo‑1B model on ARC, GPQA Diamond, and MMLU STEM benchmarks, achieving up to +28.57% improvement on ARC‑E and a 99.45% valid answer rate.

By Davide Vitabile, N. Ranjan, Akshay Nambiar, Kamal K. Gupta, Amril Nazir
llmsefficiencybenchmarks
More like this →