Xmodel-LM Technical Report

TL;DR

Xmodel-LM is a compact 1.1B language model trained on a large, balanced dataset, achieving superior performance compared to similar-sized open-source models, with accessible code and checkpoints.

Contribution

The paper introduces Xmodel-LM, a smaller yet effective language model trained on a large, balanced dataset, outperforming comparable open-source models.

Findings

Surpasses existing open-source models of similar scale

Trained on 2 trillion tokens with balanced Chinese and English data

Publicly available checkpoints and code

Abstract

We introduce Xmodel-LM, a compact and efficient 1.1B language model pre-trained on around 2 trillion tokens. Trained on our self-built dataset (Xdata), which balances Chinese and English corpora based on downstream task optimization, Xmodel-LM exhibits remarkable performance despite its smaller size. It notably surpasses existing open-source language models of similar scale. Our model checkpoints and code are publicly accessible on GitHub at https://github.com/XiaoduoAILab/XmodelLM.