Windows GPU installation error

[oddstan]

Hello.
I have same error #20, i guess it was not solved:

was installing docker gpu version.

at the beginning it gives error on downloading two files:

✘ flower Error pull access denied for marker-api-gpu-image, repository does not exist or may require 'docker login' 0.9s
✘ app Error pull access denied for marker-api-gpu-image, repository does not exist or may require 'docker login'

these files are not in repository though are referenced many times in docker-compose.gpu.yml, but it also downloads Dockerfile.gpu.distributed-server. so part that downloads marker-api-gpu-image must be leftovers of old code?

Later it ends with error: KeyError: 'sdpa'

35.48     self.self_attn = MBART_ATTENTION_CLASSES[config._attn_implementation](
35.48 KeyError: 'sdpa'

which i could nt find where does it ger this parameter from.

 => ERROR [celery_worker 7/7] RUN python -c 'from marker.models import load_all_models; load_all_models()'                                                                                     36.3s
------
 > [celery_worker 7/7] RUN python -c 'from marker.models import load_all_models; load_all_models()':
35.47 Traceback (most recent call last):
35.47   File "<string>", line 1, in <module>
35.47   File "/usr/local/lib/python3.10/dist-packages/marker/models.py", line 74, in load_all_models
35.47     order = setup_order_model(device, dtype)
35.47   File "/usr/local/lib/python3.10/dist-packages/marker/models.py", line 61, in setup_order_model
35.47     model = load_order_model()
35.47   File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/model.py", line 27, in load_model
35.47     model = OrderVisionEncoderDecoderModel.from_pretrained(checkpoint, config=config, torch_dtype=dtype)
35.47   File "/usr/local/lib/python3.10/dist-packages/transformers/models/vision_encoder_decoder/modeling_vision_encoder_decoder.py", line 376, in from_pretrained
35.47     return super().from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
35.47   File "/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py", line 4111, in from_pretrained
35.47     model = cls(config, *model_args, **model_kwargs)
35.47   File "/usr/local/lib/python3.10/dist-packages/transformers/models/vision_encoder_decoder/modeling_vision_encoder_decoder.py", line 199, in __init__
35.47     decoder = AutoModelForCausalLM.from_config(config.decoder)
35.47   File "/usr/local/lib/python3.10/dist-packages/transformers/models/auto/auto_factory.py", line 440, in from_config
35.47     return model_class._from_config(config, **kwargs)
35.47   File "/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py", line 1451, in _from_config
35.47     model = cls(config, **kwargs)
35.47   File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 495, in __init__
35.48     self.model = MBartOrderDecoderWrapper(config)
35.48   File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 480, in __init__
35.48     self.decoder = MBartOrderDecoder(config)
35.48   File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 294, in __init__
35.48     self.layers = nn.ModuleList([MBartOrderDecoderLayer(config) for _ in range(config.decoder_layers)])
35.48   File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 294, in <listcomp>
35.48     self.layers = nn.ModuleList([MBartOrderDecoderLayer(config) for _ in range(config.decoder_layers)])
35.48   File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 209, in __init__
35.48     self.self_attn = MBART_ATTENTION_CLASSES[config._attn_implementation](
35.48 KeyError: 'sdpa'
35.48 Loaded detection model vikp/surya_det3 on device cpu with dtype torch.float32
35.48 Loaded detection model vikp/surya_layout3 on device cpu with dtype torch.float32
------
failed to solve: process "/bin/sh -c python -c 'from marker.models import load_all_models; load_all_models()'" did not complete successfully: exit code: 1

i wonder if this is package bug at all.
is this key defined in source code?

if i get it right for some reason transformers pass sdpa to surya.
File "/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py", line 1451, in _from_config
35.71 model = cls(config, **kwargs)
35.71 File "/usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 495, in init
35.71 self.model = MBartOrderDecoderWrapper(config)

i tried both
pip install transformers==4.41.0
pip install transformers==4.45.2

i guess it has to do something to do with usr/local/lib/python3.10/dist-packages/surya/model/ordering/decoder.py", line 209,

same error was here wisupai/e2m#13, completely different package
and that helped:
"Same Question tring add config._attn_implementation = "eager" to miniforge/base/envs/e2m/lib/python3.10/site-packages/surya/model/ordering/decoder.py Line 208 may be work."

but thats in surya's part which cant be modified if installing subj as docker container. or i'm wrong?