AI Paper Atlas

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

Beyond Log Likelihood: Probability-Based Objectives for Supervised Fine-Tuning across the Model Capability Continuum

Gaotang Li, Ruizhong Qiu, Xiusi Chen, Heng Ji, Hanghang Tong

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 0 citations (OpenAlex)

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
replace -log(p) with -p objective in model-strong math SFTQwen2.5-Math-1.5B SFT · Math avg (5 benchmarks)accuracy: 17.00 -> 32.75Improved
exclude tokens with p<0.2 from -log(p) loss (ablation)Qwen2.5-Math-7B SFT · Math avg (5 benchmarks)accuracy: 22.67 -> 38.16Improved
replace -log(p) with -p on medical reasoning (model-intermediate)SFT on m23k · Medical reasoning avg (10 benchmarks)accuracy: LLaMA-3.1-8B 47.23 -> 45.89; Qwen2.5-Math-7B 33.56 -> 33.83No clear effect
replace -log(p) with -p on figfont puzzles (model-weak)Qwen2.5-7B SFT · Figfont puzzlesJaro-Winkler similarity: 82.48 -> 10.15Hurt
replace -log(p) with -p on low-resource multilingual instruction tuningLLaMA-3.1-8B SFT · MMLU-ProX low-resource languages avgaccuracy: 16.1 -> 11.5Hurt
replace -log(p) with -p as backbone grows from 3B to 14B in general instruction tuningQwen2.5 SFT · AlpacaEval2 LC win rate of -p vs -log pwin rate: 41.0 -> 49.6 -> 57.5Mixed
replace -log(p) with -p for coding SFTQwen2.5-Coder-7B · EvalPlus avgpass@1: 74.7 -> 77.8Improved
scale backbone from 3B to 32B in math SFT with -p versus NLL (ablation)Qwen2.5 SFT · Math benchmarksperformance improvement of -p over NLLImproved
train only on top-10% highest-probability tokens (instead of all tokens) (ablation)Qwen2.5-Math-1.5B SFT, all objectives · Math avgaccuracyImproved
include low-probability tokens in training (ablation)Qwen2.5-Math-1.5B SFT · Math avgaccuracyHurt
increase alpha (more concave objective) in model-strong end (ablation)Qwen2.5-Math-1.5B, f_alpha family · Math mean accuracyaccuracyImproved
increase alpha toward 1 (more concave objective) in model-weak end (ablation)Qwen2.5-7B, f_alpha family · Figfont Jaro-Winkler similarityJaro-Winkler similarityHurt

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