Use flash attention with VLM

### Question

I am trying to execute https://docling-project.github.io/docling/examples/compare_vlm_models/ with flash attention enabled:

```python
pipeline_options.accelerator_options.cuda_use_flash_attention2 = True
```

This fails with:

```

FlashAttention only support fp16 and bf16 data type You are attempting to use Flash Attention 2.0 without specifying a torch dtype. This might lead to unexpected behaviour docling see when enabling

```

What do I need to do to properly enable flash attention?