AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

LLM-JEPA: Large Language Models Meet Joint Embedding Predictive Architectures

Hai Huang, Yann LeCun, Randall Balestriero

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Multilingual LLM data and culture · 29 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
add JEPA loss (lambda=1,k=1) to next-token fine-tuningLlama-3.2-1B-Instruct · NL-RX-SYNTHaccuracy: 57.29 -> 71.46Improved
add JEPA loss to fine-tuning of Llama-3.1-8BLlama-3.1-8B-Instruct · NL-RX-SYNTHaccuracy: 35.77 -> 63.57Improved
add JEPA loss to fine-tuning of OLMo-2-0425-1BOLMo-2-0425-1B-Instruct · NL-RX-SYNTHaccuracy: 87.09 -> 87.52No clear effect
add JEPA loss to pretraining from scratchLlama-3.2-1B-Instruct · NL-RX-SYNTH pretrainingaccuracy: 54.38 -> 60.59Improved
add JEPA loss to fine-tuning on GSM8KLlama-3.2-1B-Instruct · GSM8Kaccuracy: 32.36 -> 36.36Improved
add JEPA loss to fine-tuning on NQ-OpenLlama-3.2-1B-Instruct · NQ-Openaccuracy: 20.12 -> 21.59Improved
replace cosine similarity with l2-norm in JEPA loss (ablation)LLM-JEPA · NL-RX-SYNTHaccuracy: 71.46 -> 2.22Hurt
replace cosine similarity with MSE (ablation)LLM-JEPA · NL-RX-SYNTHaccuracy: 71.46 -> 70.64Hurt
append predictor token instead of prepending (ablation)LLM-JEPA · NL-RX-SYNTHaccuracy: 68.07 -> 71.46Improved
predict Code from Text instead of Text from Code (ablation)LLM-JEPA · NL-RX-SYNTHaccuracy: 65.70 -> 71.46Improved
use cosine JEPA loss instead of InfoNCE (ablation)LLM-JEPA · NL-RX-SYNTHaccuracy: 34.40 -> 71.46Improved
add loss dropout LD=0.75 at matched compute (first row) (ablation)LLM-JEPA · NL-RX-SYNTHaccuracy: 19.85 -> 32.46Improved

Later papers that compared with this method

1 later accepted paper reports LLM-JEPA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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