use_amp or apex not effective

[IcanDoItL]

The RTX 3080 Ti is set with half_precision_backend: 'auto' and fp16=True, but it’s not effective and memory usage is not reduced. What could be the possible reasons?

[IcanDoItL]

args = SentenceTransformerTrainingArguments(
    # Required parameter:
    output_dir=output_dir,
    # Optional training parameters:
    num_train_epochs=num_train_epochs,
    per_device_train_batch_size=batch_size,
    per_device_eval_batch_size=batch_size,
    warmup_ratio=0.1,
    fp16=True,  # Set to False if you get an error that your GPU can't run on FP16
    bf16=False,  # Set to True if you have a GPU that supports BF16
    batch_sampler=BatchSamplers.NO_DUPLICATES,  # MultipleNegativesRankingLoss benefits from no duplicate samples in a batch
    multi_dataset_batch_sampler=MultiDatasetBatchSamplers.PROPORTIONAL,  # PROPORTIONAL or ROUND_ROBIN
    # Optional tracking/debugging parameters:
    eval_strategy="steps",
    eval_steps=250,
    save_strategy="steps",
    save_steps=250,
    save_total_limit=2,
    logging_steps=100,
    run_name="mnrl-cl-multi",  # Will be used in W&B if `wandb` is installed
)

platform: centos 7 pytorch==2.2.1 pytorch-cuda==12.1