Fix OutOfMemoryError

✅ Solution

OutOfMemoryError in transformers usually stems from excessively large models or batch sizes exceeding available GPU memory. Reduce the batch size during training/inference, enable gradient accumulation, or explore model parallelism techniques like `accelerate` or `torch.distributed.nn.Module` to distribute the model across multiple devices. Consider using quantization techniques (e.g., bitsandbytes) or offloading layers to CPU/disk (using accelerate's device_map) to further decrease memory footprint.