LiLiuM: eBay's Large Language Models for e-commerce

TL;DR

eBay developed the LiLiuM series of large language models tailored for e-commerce, achieving competitive performance on general benchmarks and superior results on e-commerce and multilingual tasks, with full control over data and architecture.

Contribution

The paper presents eBay's in-house developed LiLiuM LLMs, optimized for e-commerce, including custom training data, tokenizer, and architecture, outperforming external models on specific tasks.

Findings

LiLiuM models perform on par with LLaMA-2 on English NLU benchmarks.

LiLiuM outperforms LLaMA-2 on non-English NLU and e-commerce tasks.

Checkpoint averaging improves model performance.

Abstract

We introduce the LiLiuM series of large language models (LLMs): 1B, 7B, and 13B parameter models developed 100% in-house to fit eBay's specific needs in the e-commerce domain. This gives eBay full control over all aspects of the models including license, data, vocabulary, and architecture. We expect these models to be used as a foundation for fine-tuning and instruction-tuning, eliminating dependencies to external models. The LiLiuM LLMs have been trained on 3 trillion tokens of multilingual text from general and e-commerce domain. They perform similar to the popular LLaMA-2 models on English natural language understanding (NLU) benchmarks. At the same time, we outperform LLaMA-2 on non-English NLU tasks, machine translation and on e-commerce specific downstream tasks. As part of our data mixture, we utilize the newly released RedPajama-V2 dataset for training and share our insights regarding data filtering and deduplication. We also discuss in detail how to serialize structured data for use in autoregressive language modeling. We provide insights on the effects of including code and parallel machine translation data in pre-training. Furthermore, we develop our own tokenizer and model vocabulary, customized towards e-commerce. This way, we can achieve up to 34% speed-up in text generation on eBay-specific downstream tasks compared to LLaMA-2. Finally, in relation to LLM pretraining, we show that checkpoint averaging can further improve over the best individual model checkpoint.