Adding Space-Time-MiniLM-v0 #39479
Description
Model description
The model is an encoder, used for sentence embedding. It uses a custom attention mechanism to incorporate space-time information into the embeddings. This would be very helpful especially in retrieval scenarios.
This has not been done before, and it is not available yet to be used. We aim to have this the start of a series of models that do this task.
The All-MiniLM-L2-v6 is the closest since our model was built on top of this one's knowledge.
It uses the same tokenizer as MiniLM.
Open source status
- The model implementation is available
- The model weights are available
Provide useful links for the implementation
No response