ELO: Efficient Layer-Specific Optimization for Continual Pretraining of Multilingual LLMs

• URL: http://arxiv.org/abs/2601.03648v1
• Date: Wed, 07 Jan 2026 06:55:29 GMT
• Title: ELO: Efficient Layer-Specific Optimization for Continual Pretraining of Multilingual LLMs
• Authors: HanGyeol Yoo, ChangSu Choi, Minjun Kim, Seohyun Song, SeungWoo Song, Inho Won, Jongyoul Park, Cheoneum Park, KyungTae Lim,
• Abstract summary: We propose an efficient layer-specific optimization (ELO) method designed to enhance continual pretraining (CP) for specific languages in multilingual large language models (MLLMs)<n>The ELO method consists of two main stages: (1) ELO Pretraining, where a small subset of specific layers, identified in our experiments as the critically important first and last layers, are detached from the original MLLM and trained with the target language.<n> Experimental results demonstrate that the ELO method achieves a training speedup of up to 6.46 times compared to existing methods, while improving target language performance by up to 6.2% on qualitative benchmarks and
• Score: 10.015119836510332
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We propose an efficient layer-specific optimization (ELO) method designed to enhance continual pretraining (CP) for specific languages in multilingual large language models (MLLMs). This approach addresses the common challenges of high computational cost and degradation of source language performance associated with traditional CP. The ELO method consists of two main stages: (1) ELO Pretraining, where a small subset of specific layers, identified in our experiments as the critically important first and last layers, are detached from the original MLLM and trained with the target language. This significantly reduces not only the number of trainable parameters but also the total parameters computed during the forward pass, minimizing GPU memory consumption and accelerating the training process. (2) Layer Alignment, where the newly trained layers are reintegrated into the original model, followed by a brief full fine-tuning step on a small dataset to align the parameters. Experimental results demonstrate that the ELO method achieves a training speedup of up to 6.46 times compared to existing methods, while improving target language performance by up to 6.2\% on qualitative benchmarks and effectively preserving source language (English) capabilities.

Related papers

• Adapting Where It Matters: Depth-Aware Adaptation for Efficient Multilingual Speech Recognition in Low-Resource Languages [11.808922632545874]
  We analyze multilingual automatic speech recognition models and reveal a U-shaped adaptability pattern.<n>We propose DAMA, a Depth-Aware Model Adaptation framework that allocates adaptation capacity according to each layer's role.<n>Dama matches or surpasses state-of-the-art accuracy with 80% fewer trainable parameters, achieves a 29% error reduction under extreme data scarcity, and significantly improves memory, training time, and computational efficiency over baselines.
  arXiv  Detail & Related papers  (2026-02-01T04:18:31Z)
• SCOPE: Language Models as One-Time Teacher for Hierarchical Planning in Text Environments [4.375012768093524]
  Long-term planning in text-based environments presents significant challenges due to open-ended action spaces, ambiguous observations, and sparse feedback.<n>Recent research suggests that large language models (LLMs) encode rich semantic knowledge about the world, which can be valuable for guiding agents in high-level reasoning and planning across both embodied and purely textual settings.<n>Existing approaches often depend heavily on querying LLMs during training and inference, making them computationally expensive and difficult to deploy efficiently.<n>We introduce SCOPE (Subgoal-COnditioned Pretraining for Efficient planning), a one-shot hierarchical planner that leverages LLM-generated subgoal
  arXiv  Detail & Related papers  (2025-12-10T18:26:14Z)
• Language Ranker: A Lightweight Ranking framework for LLM Decoding [70.01564145836129]
  This paper conceptualizes the decoding process as analogous to the ranking stage in recommendation pipelines.<n>Motivated by this insight, we propose Language Ranker, a novel framework that introduces a lightweight module to rerank candidate responses.<n> Experiments show that Language Ranker achieves performance comparable to large-scale reward models, while requiring only 0.5M additional parameters.
  arXiv  Detail & Related papers  (2025-10-23T17:56:46Z)
• Enhancing Non-English Capabilities of English-Centric Large Language Models through Deep Supervision Fine-Tuning [42.166438218926274]
  We introduce a deep supervision fine-tuning method (DFT) that incorporates additional supervision in the internal layers of the model to guide its workflow.<n>Our method guides the model to not only consider the final generated result when processing non-English inputs but also ensure the accuracy of internal representations.
  arXiv  Detail & Related papers  (2025-03-03T07:59:32Z)
• SLAM: Towards Efficient Multilingual Reasoning via Selective Language Alignment [78.4550589538805]
  We propose an efficient multilingual reasoning alignment approach that precisely identifies and fine-tunes the layers responsible for handling multilingualism.<n> Experimental results show that our method, SLAM, only tunes 6 layers' feed-forward sub-layers including 6.5-8% of all parameters within 7B and 13B LLMs.
  arXiv  Detail & Related papers  (2025-01-07T10:29:43Z)
• Adaptive Pruning for Large Language Models with Structural Importance Awareness [66.2690963378878]
  Large language models (LLMs) have significantly improved language understanding and generation capabilities.<n>LLMs are difficult to deploy on resource-constrained edge devices due to their high computational and storage resource demands.<n>We propose structurally-aware adaptive pruning (SAAP) to significantly reduce the computational and memory costs while maintaining model performance.
  arXiv  Detail & Related papers  (2024-12-19T18:08:04Z)
• Optimizing Low-Resource Language Model Training: Comprehensive Analysis of Multi-Epoch, Multi-Lingual, and Two-Stage Approaches [3.809045695573932]
  Existing works adopt multi-epoch, multi-lingual, and two-stage training to utilize the limited target language corpus efficiently. We exhaustively explore training setups for low-resource language LLM, combining these three approaches. As the amount of target language corpus decreases, the optimal training approach shifts from monolingual single-stage training to multi-lingual two-stage training at a compute budget dependent threshold.
  arXiv  Detail & Related papers  (2024-10-16T07:45:56Z)
• Exploring Design Choices for Building Language-Specific LLMs [36.32622880071991]
  We study building language-specific language models by adapting monolingual and multilingual models. We find that the initial performance of LLM does not always correlate with the final performance after the adaptation.
  arXiv  Detail & Related papers  (2024-06-20T18:47:43Z)
• Analyzing and Adapting Large Language Models for Few-Shot Multilingual NLU: Are We There Yet? [82.02076369811402]
  Supervised fine-tuning (SFT), supervised instruction tuning (SIT) and in-context learning (ICL) are three alternative, de facto standard approaches to few-shot learning. We present an extensive and systematic comparison of the three approaches, testing them on 6 high- and low-resource languages, three different NLU tasks, and a myriad of language and domain setups. Our observations show that supervised instruction tuning has the best trade-off between performance and resource requirements.
  arXiv  Detail & Related papers  (2024-03-04T10:48:13Z)
• Accelerating LLaMA Inference by Enabling Intermediate Layer Decoding via Instruction Tuning with LITE [62.13435256279566]
  Large Language Models (LLMs) have achieved remarkable performance across a wide variety of natural language tasks. However, their large size makes their inference slow and computationally expensive. We show that it enables these layers to acquire 'good' generation ability without affecting the generation ability of the final layer.
  arXiv  Detail & Related papers  (2023-10-28T04:07:58Z)
• Bridging the Gap between Language Models and Cross-Lingual Sequence Labeling [101.74165219364264]
  Large-scale cross-lingual pre-trained language models (xPLMs) have shown effectiveness in cross-lingual sequence labeling tasks. Despite the great success, we draw an empirical observation that there is a training objective gap between pre-training and fine-tuning stages. In this paper, we first design a pre-training task tailored for xSL named Cross-lingual Language Informative Span Masking (CLISM) to eliminate the objective gap. Second, we present ContrAstive-Consistency Regularization (CACR), which utilizes contrastive learning to encourage the consistency between representations of input parallel
  arXiv  Detail & Related papers  (2022-04-11T15:55:20Z)
• Mixed-Lingual Pre-training for Cross-lingual Summarization [54.4823498438831]
  Cross-lingual Summarization aims at producing a summary in the target language for an article in the source language. We propose a solution based on mixed-lingual pre-training that leverages both cross-lingual tasks like translation and monolingual tasks like masked language models. Our model achieves an improvement of 2.82 (English to Chinese) and 1.15 (Chinese to English) ROUGE-1 scores over state-of-the-art results.
  arXiv  Detail & Related papers  (2020-10-18T00:21:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.