Why is flash attention not supported when fine tuning ovis2

[mehamednews]

Describe the bug
I tried running ms-swift to fine tune ovis2 (https://huggingface.co/AIDC-AI/Ovis2-1B).
I keep getting an error that ovis2 does not support flash attetion but they claim to support flash attention in their github repository: AIDC-AI/Ovis#44

without flash attention I always get OOM errors (even with the 1B model on a 4090).
this is the command I'm running:

USE_HF=1 CUDA_VISIBLE_DEVICES=0 swift sft \
  --model ovis2 \
  --model AIDC-AI/Ovis2-1B \
  --output-dir ovis2-1b-att1 \
  --train_type lora \
  --freeze_vit true \
  --lora_rank 8 \
  --lora_alpha 32 \
  --dataset ./label-dataset-train.jsonl \
  --val_dataset ./label-dataset-val.jsonl \
  --num_train_epochs 3 \
  --learning_rate 5e-6 \
  --warmup_ratio 0.25 \
  --save_only_model true \
  --eval_steps 25 \
  --dataloader_num_workers 1 \
  --max_pixels 1003520 \
  --attn_impl flash_attn

ValueError: Ovis does not support Flash Attention 2.0 yet. Please request to add support where the model is hosted, on its model hub page: https://huggingface.co//home/ab/.cache/huggingface/hub/models--AIDC-AI--Ovis2-1B/snapshots/2d2b94c33f1876797bead75c057cf9a9d6e2181a/discussions/new or in the Transformers GitHub repo: https://github.com/huggingface/transformers/issues/new

I'm not sure if I'm doing something wrong. Any help would be appreciated.