Serving or create model manually

you can create model and start batch transform job in one command.

Run Locally

install dependencies run python debug/batch_manager.py

For model creation and start TransformJob on GPU

• default model is sentence-transformers/LaBSE to change the model send - all-MiniLM-L6-v2 use this commands:

export MODEL_ARTIFACTS=s3://llama-weights-us/labse-model/model.tar.gz

python create_model_and_inference.py \
--sg_model_name all-mini \ 
--model_artifact_path $MODEL_ARTIFACTS \
--batch_size 64 \
--strategy MultiRecord \
--create_model \
--instance_type  ml.g4dn.xlarge  \
--input_data s3://llama-weights-us/input_data/ \
--output_data s3://llama-weights-us/output_data/ \
--st_model all-MiniLM-L12-v2 \
--serve 

For creating batch transform job manully use this command:

python create_model_and_inference.py \
--model_name all-mini \
--input_data s3://llama-weights-us/input_data/  \
--output_data s3://llama-weights-us/output_data/ \
--serve