Error during trying to run train script

[LeonGoretzkatju]

Hi,

Thank you so much for open-sourcing this wonderful VLM-based VLN framework, during the reproduction of your training part, I meet one problem,

MID_RUN_NAME: StreamVLN_Video_qwen_1_5_1epoch_196token_8history_32frame
[2025-07-28 21:14:16,260] torch.distributed.run: [WARNING] master_addr is only used for static rdzv_backend and when rdzv_endpoint is not specified.
[2025-07-28 21:14:16,260] torch.distributed.run: [WARNING]
[2025-07-28 21:14:16,260] torch.distributed.run: [WARNING] *****************************************
[2025-07-28 21:14:16,260] torch.distributed.run: [WARNING] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
[2025-07-28 21:14:16,260] torch.distributed.run: [WARNING] *****************************************
[2025-07-28 21:14:32,671] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2025-07-28 21:14:32,705] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[WARNING] async_io requires the dev libaio .so object and headers but these were not found.
[WARNING] async_io requires the dev libaio .so object and headers but these were not found.
[WARNING] async_io: please install the libaio-dev package with apt
[WARNING] If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
[WARNING] Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH
[WARNING] async_io: please install the libaio-dev package with apt
[WARNING] If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
[WARNING] Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH
[WARNING] sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.1
[WARNING] using untested triton version (2.1.0), only 1.0.0 is known to be compatible
[WARNING] sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.1
[WARNING] using untested triton version (2.1.0), only 1.0.0 is known to be compatible
/hpc2hdd/home/zzhong778/.conda/envs/streamvln/lib/python3.9/site-packages/transformers/training_args.py:1545: FutureWarning: evaluation_strategy is deprecated and will be removed in version 4.46 of 🤗 Transformers. Use eval_strategy instead

The error message is:
ImportError: FlashAttention2 has been toggled on, but it cannot be used due to the following error: the package flash_attn seems to be not installed. Please refer to the documentation of https://huggingface.co/docs/transformers/perf_infer_gpu_one#flashattention-2 to install Flash Attention 2.

Since I install the required library of stream-vln using your provided requirements.txt, after checking this txt file, I didn't find the relative part to install falsh_attention explicitly, could you please provide more information about configuring the streamvln train environment?

Thanks a lot!

[kellyiss]

You can try pip install flash-attn==2.5.8 --no-build-isolation

[LeonGoretzkatju]

You can try pip install flash-attn==2.5.8 --no-build-isolation

Thank you for your kind reply, now I have solved this problem and successfully trained your model. Thank you!