This is the training curves of model Mantis-8B-SigLIP-LLaMA-3. It's reported to help reproduce mantis's training results. This training has 1 training failure, and the model trainings is resumed from the previous checkpoint with the same hyper parameters.

This is the training curves of model Mantis-8B-CLIP-LLaMA-3. It's reported to help reproduce mantis's training results. This training has 2 training failures, and models trainings are resumed from the previous checkpoint with the same hyper parameters