Embedding model to inference mode (llm 01L)

[patputnam-db]

The TransformerEncoder model in CMD 14 is instantiated in 'training' mode. As such, the dropout layers are not frozen during inference. As such, re-running the embedding model multiple times on the same input tensors results in different embeddings.

Fix: Add a model.eval() call after the model is instantiated (cmd 14, line 11).