Training large models requires a large amount of data, as well as abundant computation resources. While collaborative learning provides a promising paradigm to harness collective data from many participants, training large models remains a major challenge for participants with limited resources. We introduce MSfusion, an effective and efficient collaborative learning framework, tailored for training larger models on resource-constraint machines through model splitting. Specifically, a double shifting model splitting scheme is designed. Each participant is assigned a subset of model parameters to train over local data, and aggregates with sub-models of other peers on common parameters. While model splitting significantly reduces the computation and communication costs of individual participants, additional novel designs on adaptive model overlapping and contrastive loss functions help MSfusion to maintain training effectiveness, against model shift across participants. Extensive experiments on image and NLP tasks illustrate significant advantages of MSfusion in performance and efficiency for training large models, and its strong scalability: computation cost of each participant reduces significantly as the number of participants increases.