AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

SPR-RAFT: Parameter-Efficient Regression-Aware Fine-Tuning for Biomedical LLM Regression

Yuanlin Yang, Chenhui Li, Xuhao Guo, Anqi Zhang, Hoi Leong Lee, Haodong Liu

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add GLS correlation-aware fusion instead of naive IVW (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.73 -> 0.70Improved
add regression head to soft prompt (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.78 -> 0.70Improved
add soft prompt to regression head (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.74 -> 0.70Improved
lengthen soft prompt from 16 to 32 (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.72 -> 0.70Improved
lengthen soft prompt from 32 to 64 (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.70 -> 0.69No clear effect
swap Qwen3-4B for Llama-3.2-3B backbone (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.70 -> 0.93Hurt
swap Qwen3-4B for Gemma3-4B backbone (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.70 -> 0.80Hurt
scale backbone from Qwen3-4B to Qwen3-8B (ablation)SPR-RAFT · TrialBench Duration Phase IMAE: 0.70 -> 0.68No clear effect
use SPR-RAFT instead of RAFT on Llama-3.2-3Bbackbone · TrialBench Duration Phase IMAE: 1.92 -> 0.93Improved
use SPR-RAFT instead of RAFT on Qwen3-8Bbackbone · TrialBench Duration Phase IMAE: 0.88 -> 0.68Improved
fuse head and text branches instead of text-only (ablation)SPR-RAFT · uncertainty diagnosticMSE: 1.345 -> 1.129Improved
use SPR-RAFT instead of RAFTLLM regression · NHANES (2007-2018)AUC: 0.75 -> 0.84Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy