SentenceTransformerTrainer: checkpoint saves custom_st.py in the wrong folder
When you train with arguments
save_strategy="steps"
save_steps=100
the checkpoint saves custom_st.py in subfolder 0_Transformer, but it should be saved in the parent folder, I think.
Otherwise when loading the checkpoint, you get an error:ModuleNotFoundError: No module named 'custom_st'
sentence-transformers 3.2.0
transformers 4.45.2
Or maybe I should load from the 0_Transformer subfolder instead of the checkpoint folder?
Because previously mentioned workaround does not load the weights.
Hi
@botkop
, are you loading the checkpoint to continue fine-tuning or just for inference? If it's for inference, you can replace the model weights in jinaai/jina-embeddings-v3 with the weights from the checkpoint directory and use the fine-tuned model that way
Yes, it's for inference after fine-tuning.
With model weights, you mean checkpoint-1200/0_Transformer/model.safetensors, correct?
Yes, correct