Machine Translation Evaluation Metrics

When using the Unbabel/wmt20-comet-qe-da metric, do the approaches below have any effect on evaluation: (1) Evaluating each model translation separately or (2) in the same data list? For example, when translating the English word "Hello" to Afrikaans: (1) Evaluating each model translation separately data1 = [ { "src": "hello", "mt": "hallllo" }, ] model_output1 = model.predict(data1, batch_size=8, gpus=1) print (model_output1) data2 = [ { "src": "hello", "mt": "hallo" } ] model_output2 = model.predict(data2, batch_size=8, gpus=1) print (model_output2) (2) in the same data list data = [ { "src": "hello", "mt": "hallo" }, { "src": "hello", "mt": "hallllo" }, ] model_output = model.predict(data, batch_size=8, gpus=1) print (model_output)