KeyError: 'sdpa'

[miloskovacevic68]

I installed the newest version of marker-pdf with pip install marker-pdf in a newly created venv environment with python 3.10.
When i try to use marker_single from cli i get the error:

Loaded detection model vikp/surya_det3 on device cuda with dtype torch.float16
Loaded detection model vikp/surya_layout3 on device cuda with dtype torch.float16
Traceback (most recent call last):
File "/home/milos/PycharmProjects/raga_marker/venv/bin/marker_single", line 8, in
sys.exit(main())
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/convert_single.py", line 31, in main
model_lst = load_all_models()
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/marker/models.py", line 78, in load_all_models
order = setup_order_model(device, dtype)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/marker/models.py", line 66, in setup_order_model
model = load_order_model()
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/surya/model/ordering/model.py", line 27, in load_model
model = OrderVisionEncoderDecoderModel.from_pretrained(checkpoint, config=config, torch_dtype=dtype)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/transformers/models/vision_encoder_decoder/modeling_vision_encoder_decoder.py", line 376, in from_pretrained
return super().from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 4096, in from_pretrained
model = cls(config, *model_args, **model_kwargs)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/transformers/models/vision_encoder_decoder/modeling_vision_encoder_decoder.py", line 199, in init
decoder = AutoModelForCausalLM.from_config(config.decoder)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 440, in from_config
return model_class._from_config(config, **kwargs)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 1543, in _from_config
model = cls(config, **kwargs)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/surya/model/ordering/decoder.py", line 495, in init
self.model = MBartOrderDecoderWrapper(config)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/surya/model/ordering/decoder.py", line 480, in init
self.decoder = MBartOrderDecoder(config)
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/surya/model/ordering/decoder.py", line 294, in init
self.layers = nn.ModuleList([MBartOrderDecoderLayer(config) for _ in range(config.decoder_layers)])
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/surya/model/ordering/decoder.py", line 294, in
self.layers = nn.ModuleList([MBartOrderDecoderLayer(config) for _ in range(config.decoder_layers)])
File "/home/milos/PycharmProjects/raga_marker/venv/lib/python3.10/site-packages/surya/model/ordering/decoder.py", line 209, in init
self.self_attn = MBART_ATTENTION_CLASSES[config._attn_implementation](
KeyError: 'sdpa'

Please help!

[AdrianY0809]

The same problem

[xunmenglt]

I solve it by 'pip install transformers==4.45.2'

[MarcinRutecki]

@xunmenglt thank you. It worked

[johmicrot]

I also confirm @xunmenglt solved the issue for me.

[VikParuchuri]

Yes, this is a bug with transformers 4.46. Will patch today

[Jensssen]

This is what I call realtime-bug-fixing. Amazing! 🥇

[VikParuchuri]

See #319 and VikParuchuri/surya#226, will merge shortly

[miloskovacevic68]

Thank you. It works well!

…

See #319, will merge shortly -- Reply to this email directly or view it on GitHub: #318 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

[WaleedChughtai30]

I solve it by 'pip install transformers==4.45.2'

worked for me too!

[bineea]

I solve it by 'pip install transformers==4.45.2'

worked for me too!