MobileLLM-Pro Technical Report

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Huber, Patrick, Chang, Ernie, Wen, Wei, Fedorov, Igor, Elgamal, Tarek, Huang, Hanxian, Suda, Naveen, Sankar, Chinnadhurai, Vogeti, Vish, Wang, Yanghan, Gladkov, Alex, Tai, Kai Sheng, Elogeel, Abdelrahman, Hefny, Tarek, Chandra, Vikas, Aly, Ahmed, Kumar, Anuj, Krishnamoorthi, Raghuraman, Sagar, Adithya
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908640404832256
author Huber, Patrick
Chang, Ernie
Wen, Wei
Fedorov, Igor
Elgamal, Tarek
Huang, Hanxian
Suda, Naveen
Sankar, Chinnadhurai
Vogeti, Vish
Wang, Yanghan
Gladkov, Alex
Tai, Kai Sheng
Elogeel, Abdelrahman
Hefny, Tarek
Chandra, Vikas
Aly, Ahmed
Kumar, Anuj
Krishnamoorthi, Raghuraman
Sagar, Adithya
author_facet Huber, Patrick
Chang, Ernie
Wen, Wei
Fedorov, Igor
Elgamal, Tarek
Huang, Hanxian
Suda, Naveen
Sankar, Chinnadhurai
Vogeti, Vish
Wang, Yanghan
Gladkov, Alex
Tai, Kai Sheng
Elogeel, Abdelrahman
Hefny, Tarek
Chandra, Vikas
Aly, Ahmed
Kumar, Anuj
Krishnamoorthi, Raghuraman
Sagar, Adithya
contents Efficient on-device language models around 1 billion parameters are essential for powering low-latency AI applications on mobile and wearable devices. However, achieving strong performance in this model class, while supporting long context windows and practical deployment remains a significant challenge. We introduce MobileLLM-Pro, a 1-billion-parameter language model optimized for on-device deployment. MobileLLM-Pro achieves state-of-the-art results across 11 standard benchmarks, significantly outperforming both Gemma 3-1B and Llama 3.2-1B, while supporting context windows of up to 128,000 tokens and showing only minor performance regressions at 4-bit quantization. These improvements are enabled by four core innovations: (1) implicit positional distillation, a novel technique that effectively instills long-context capabilities through knowledge distillation; (2) a specialist model merging framework that fuses multiple domain experts into a compact model without parameter growth; (3) simulation-driven data mixing using utility estimation; and (4) 4-bit quantization-aware training with self-distillation. We release our model weights and code to support future research in efficient on-device language models.
format Preprint
id arxiv_https___arxiv_org_abs_2511_06719
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MobileLLM-Pro Technical Report
Huber, Patrick
Chang, Ernie
Wen, Wei
Fedorov, Igor
Elgamal, Tarek
Huang, Hanxian
Suda, Naveen
Sankar, Chinnadhurai
Vogeti, Vish
Wang, Yanghan
Gladkov, Alex
Tai, Kai Sheng
Elogeel, Abdelrahman
Hefny, Tarek
Chandra, Vikas
Aly, Ahmed
Kumar, Anuj
Krishnamoorthi, Raghuraman
Sagar, Adithya
Machine Learning
Efficient on-device language models around 1 billion parameters are essential for powering low-latency AI applications on mobile and wearable devices. However, achieving strong performance in this model class, while supporting long context windows and practical deployment remains a significant challenge. We introduce MobileLLM-Pro, a 1-billion-parameter language model optimized for on-device deployment. MobileLLM-Pro achieves state-of-the-art results across 11 standard benchmarks, significantly outperforming both Gemma 3-1B and Llama 3.2-1B, while supporting context windows of up to 128,000 tokens and showing only minor performance regressions at 4-bit quantization. These improvements are enabled by four core innovations: (1) implicit positional distillation, a novel technique that effectively instills long-context capabilities through knowledge distillation; (2) a specialist model merging framework that fuses multiple domain experts into a compact model without parameter growth; (3) simulation-driven data mixing using utility estimation; and (4) 4-bit quantization-aware training with self-distillation. We release our model weights and code to support future research in efficient on-device language models.
title MobileLLM-Pro Technical Report
topic Machine Learning
url https://arxiv.org/abs/2511.06719