AI Paper Atlas

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

TokMem: Tokenized Procedural Memory for Large Language Models

Zijun Wu, Yongchang Hao, Lili Mou

ICLR 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
use TokMem instead of LoRA fine-tuningQwen 2.5 0.5B · SNI 1000 tasksROUGE-L: 43.2 -> 50.0Improved
use TokMem instead of LoRA fine-tuningLlama 3.1 8B · SNI 1000 tasksROUGE-L: 61.6 -> 64.8Improved
use TokMem routing instead of Sentence-BERT RAG retrievalQwen 2.5 0.5B · SNI 1000 tasks routingaccuracy: 79.7 -> 94.7Improved
add replay memory to fine-tuningLoRA fine-tuning · Qwen 2.5 0.5B SNI 1000 tasksROUGE-L: 43.2 -> 46.7Improved
decouple memory token into address and steering embeddings (ablation)TokMem · SNI average, Qwen 0.5B vs Llama 3BROUGE-L: 50.7 -> 51.1 (Qwen); 62.9 -> 62.6 (Llama 3B)No clear effect
add renormalization of new memory embeddings (ablation)TokMem · continual tool learningretention of old toolsImproved
add brief adaptation of backbone on auxiliary tools (ablation)TokMem · APIGen tool selection avg, Llama 3.2 1BF1: 86.2 -> 98.4Improved
add brief adaptation of backbone on auxiliary tools (ablation)TokMem · APIGen tool selection avg, Llama 3.1 8BF1: 82.8 -> 99.1Improved
use TokMem (w/ adapt) instead of LoRA fine-tuningLlama 3.2 1B · APIGen tool selection avgF1: 9.0 -> 98.4Improved
use TokMem (w/ adapt) instead of LoRA fine-tuningLlama 3.2 3B · APIGen argument avgF1: 85.8 -> 86.3No clear effect
use TokMem instead of RAGLlama 3.2 3B · APIGen tool selection avgF1: 72.8 -> 99.2Improved
infix memory placement instead of prefix (ablation)TokMem-1 · Fanfics 1024 tokensperplexity: 3.81 -> 3.28Improved

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