dice-group

Add a new Knowledge Graph Embedding model to the dicee framework. Use when implementing a new KGE model, adding a custom scoring function, creating a new algebra-based embedding, or extending BaseKGE with a new architecture.

Add a new Knowledge Graph Embedding model to the dicee framework. Use when: implementing a new KGE model, adding a custom scoring function, creating a new algebra-based embedding, extending BaseKGE with a new architecture.

Use a pre-trained KGE model for inference via the KGE class. Use when loading a trained model, predicting missing links (head/relation/tail), answering multi-hop EPFO queries (1p 2p 3p 2i 3i ip pi 2u up), predicting literal values, accessing raw embeddings, or deploying with the Gradio UI.

Configure and launch KGE model training in dicee. Use when training a KGE model, choosing a trainer backend, selecting a scoring technique, setting up multi-GPU/DDP/FSDP training, using weight averaging (SWA/EMA/SWAG), configuring periodic evaluation, or resuming training with continual learning.

Configure and launch KGE model training in dicee. Use when: training a KGE model, choosing a trainer backend, selecting a scoring technique, setting up multi-GPU or DDP training, using weight averaging (SWA/EMA/SWAG), configuring periodic evaluation, resuming training with continual learning.

Use a pre-trained KGE model for inference via the KGE class. Use when: loading a trained model, predicting missing links (head/relation/tail), answering multi-hop EPFO queries (1p 2p 3p 2i 3i ip pi 2u up), predicting literal values, accessing raw embeddings, deploying with the Gradio UI.

Validate and use CUDA graph capture in Megatron Bridge, including local full-iteration graphs and Transformer Engine scoped graphs for attention, MLP, and MoE modules.

Operational guide for enabling MoE expert-parallel communication overlap in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.