I train a Huggingface model with fp16=True, e.g.:

    training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    learning_rate=4e-5,
    lr_scheduler_type="cosine",
    per_device_train_batch_size=32,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
    fp16=True,
)

Is there any point in also setting fp16_full_eval=True? Or is that already implied by fp16=True?

Same question for bf16.