What AI says about eCeLLM

eCeLLM is a series of e-commerce large language models created by instruction-tuning general-purpose LLMs with ECInstruct, an open-sourced, large-scale benchmark for e-commerce instruction data. ECInstruct contains 116,528 samples across 10 e-commerce tasks in 4 categories, with both in-domain and out-of-domain test sets for unseen products. Experiments show eCeLLM substantially outperforms baselines including GPT-4 Turbo on in-domain tasks and generalizes well to unseen products and instructions, indicating strong potential as a generalist e-commerce model.