Checkpoints of connectors obtained for VLMs by using HYMA (ours) and Grid Search (gs). We denote "mlp1" as the MLP_1 setting in the paper.
Jaisidh Singh
jaisidhsingh
AI & ML interests
Deep learning
Organizations
None yet
HYMA: VLM connector checkpoints
Checkpoints of connectors obtained for VLMs by using HYMA (ours) and Grid Search (gs). We denote "mlp1" as the MLP_1 setting in the paper.
HYMA: LlaVA Alignment CC3M-558K Pre-embedded
This collection presents CC3M-558K, the data used by LlaVA for multi-modal feature alignment, pre-embedded across 9 image encoders & 3 text encoders.