AI Paper Atlas

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

Don't Force the Fit: Bounded Log-Likelihood Loss for Enhanced Reasoning in Large Language Models

Feng Zhao, Hong Zhang, Yu Yang, Ruilin Zhao, Guandong Xu

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data

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 cross-entropy with bounded log-likelihood loss log(2 - p)SFT on NuminaMath-CoT 100k · 7 math benchmarks, Qwen2.5-Math-1.5BOverall Avg. accuracy: 28.36 -> 41.56Improved
standard SFT on Qwen3-8B-BaseQwen3-8B-Base · 7 math benchmarksOverall Avg. accuracy: 38.74 -> 35.24Hurt
mask loss of tokens with probability below threshold (5%, 10%, 15% quantile) (ablation)SFT Qwen2.5-Math-1.5B · AIME, Minerva, Olympiad, AMC, MATH500, GSM8KaccuracyImproved
lower learning rate 1e-6 for SFT (ablation)SFT Qwen2.5-Math-1.5B · 5 math benchmarks avgAvg. accuracy: 31.72 -> 23.43Hurt
add gradient clipping (0.3 or 0.5) to SFT (ablation)SFT Qwen2.5-Math-1.5B · 5 math benchmarks avgAvg. accuracy: 31.72 -> 32.09 (0.3), 32.13 (0.5)No clear effect
train SFT for 2 epochs (ablation)SFT Qwen2.5-Math-7B · 5 math benchmarks avgAvg. accuracy: 40.72 -> 43.60 (BLL 51.80)Improved
reduce beta toward 0 in beta-balanced loss (complement suppression only) (ablation)BLL-Loss · GSM8K, MATH500, Olympiad, Minerva, AMCaccuracyImproved
BLL-Loss versus SFT at different training set sizes (10k-100k) (ablation)Qwen2.5-Math-1.5B · data scalingaccuracyImproved
use BLL-Loss for code SFT (oss instruct)Qwen3-4B-Base · MBPPaccuracy: 38.40 -> 61.80Improved
use BLL-Loss for general reasoning SFT (alpaca-gpt4)Qwen3-4B-Base · BBHaccuracy: 51.49 -> 65.85Improved
use GEM loss instead of SFTQwen2.5-Math-1.5B · 7 math benchmarksOverall Avg. accuracy: 18.79 (Base) -> 14.04Hurt
add NEFT noise to SFTSFT Qwen2.5-Math-7B · 7 math benchmarksOverall Avg. accuracy: 35.95 -> 35.23No clear effect

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