Transformers: XLM run_squad errors with size mismatch

Created on 18 Dec 2019 · 14Comments · Source: huggingface/transformers

🐛 Bug

Model: XLM_mlm_ende_1024

Language: English

Using: the official example scripts: run_squad.py

To Reproduce

Steps to reproduce the behavior:

install dependecies and download squad v1.1 data; pull, install transformers from github master.
run run_squad.py with the following args

--model_type xlm --model_name_or_path xlm-mlm-ende-1024 --do_train --do_eval --train_file ./squad_data/train-v1.1.json --predict_file ./squad_data/dev-v1.1.json --per_gpu_train_batch_size 12 --learning_rate 3e-5 --num_train_epochs 2.0 --max_seq_length 384 --doc_stride 128 --output_dir ./debug_xlm

Ultimate error:

size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).

Full inout ouput below.

Expected behavior: finetune xlm for squad.

Environment

OS: OpenSuse 15.0
Python version: 3.6
PyTorch version: torch.__version__ = '1.3.1+cpu'
PyTorch Transformers version (or branch): (just transformers now?) 2.2.2
Using GPU ? nope
Distributed of parallel setup ? Ummm n/a

Relates to previous issues (possibly):

I've had with XLM
Similar looking error

Additional context

```python ./transformers/examples/run_squad.py --model_type xlm --model_name_or_path xlm-mlm-ende-1024 --do_train --do_eval --train_file ./squad_data/train-v1.1.json --predict_file ./squad_data/dev-v1.1.json --per_gpu_train_batch_size 12 --learning_rate 3e-5 --num_train_epochs 2.0 --max_seq_length 384 --doc_stride 128 --output_dir ./debug_xlm
12/18/2019 15:47:23 - WARNING - __main__ - Process rank: -1, device: cpu, n_gpu: 0, distributed training: False, 16-bits training: False
12/18/2019 15:47:23 - INFO - transformers.configuration_utils - loading configuration file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-config.json from cache at /HOME/.cache/torch/transformers/8f689e7cdf34bbebea67ad44ad6a142c9c5144e5c19d989839139e0d47d1ed74.0038e5c2b48fc777632fc95c3d3422203693750b1d0845a511b3bb84ad6d8c29
12/18/2019 15:47:23 - INFO - transformers.configuration_utils - Model config {
"asm": false,
"attention_dropout": 0.1,
"bos_index": 0,
"causal": false,
"dropout": 0.1,
"emb_dim": 1024,
"embed_init_std": 0.02209708691207961,
"end_n_top": 5,
"eos_index": 1,
"finetuning_task": null,
"gelu_activation": true,
"id2label": {
"0": "LABEL_0",
"1": "LABEL_1"
},
"id2lang": {
"0": "de",
"1": "en"
},
"init_std": 0.02,
"is_decoder": false,
"is_encoder": true,
"label2id": {
"LABEL_0": 0,
"LABEL_1": 1
},
"lang2id": {
"de": 0,
"en": 1
},
"layer_norm_eps": 1e-12,
"mask_index": 5,
"max_position_embeddings": 512,
"max_vocab": -1,
"min_count": 0,
"n_heads": 8,
"n_langs": 2,
"n_layers": 6,
"num_labels": 2,
"output_attentions": false,
"output_hidden_states": false,
"output_past": true,
"pad_index": 2,
"pruned_heads": {},
"same_enc_dec": true,
"share_inout_emb": true,
"sinusoidal_embeddings": false,
"start_n_top": 5,
"summary_activation": null,
"summary_first_dropout": 0.1,
"summary_proj_to_labels": true,
"summary_type": "first",
"summary_use_proj": true,
"torchscript": false,
"unk_index": 3,
"use_bfloat16": false,
"use_lang_emb": true,
"vocab_size": 30145
}

12/18/2019 15:47:24 - INFO - transformers.tokenization_utils - loading file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-vocab.json from cache at /HOME/.cache/torch/transformers/6771b710c1daf9d51643260fdf576f6353369c3563bf0fb12176c692778dca3f.2c29a4b393decdd458e6a9744fa1d6b533212e4003a4012731d2bc2261dc35f3
12/18/2019 15:47:24 - INFO - transformers.tokenization_utils - loading file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-merges.txt from cache at /HOME/.cache/torch/transformers/85d878ffb1bc2c3395b785d10ce7fc91452780316140d7a26201d7a912483e44.42fa32826c068642fdcf24adbf3ef8158b3b81e210a3d03f3102cf5a899f92a0
12/18/2019 15:47:25 - INFO - transformers.modeling_utils - loading weights file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-pytorch_model.bin from cache at /HOME/.cache/torch/transformers/ea4c0bbee310b490decb2b608a4dbc8ed9f2e4a103dd729ce183770b0fef698b.119d74257b953e5d50d73555a430ced11b1c149a7c17583219935ec1bd37d948
12/18/2019 15:47:28 - INFO - transformers.modeling_utils - Weights of XLMForQuestionAnswering not initialized from pretrained model: ['qa_outputs.start_logits.dense.weight', 'qa_outputs.start_logits.dense.bias', 'qa_outputs.end_logits.dense_0.weight', 'qa_outputs.end_logits.dense_0.bias', 'qa_outputs.end_logits.LayerNorm.weight', 'qa_outputs.end_logits.LayerNorm.bias', 'qa_outputs.end_logits.dense_1.weight', 'qa_outputs.end_logits.dense_1.bias', 'qa_outputs.answer_class.dense_0.weight', 'qa_outputs.answer_class.dense_0.bias', 'qa_outputs.answer_class.dense_1.weight']
12/18/2019 15:47:28 - INFO - transformers.modeling_utils - Weights from pretrained model not used in XLMForQuestionAnswering: ['pred_layer.proj.weight', 'pred_layer.proj.bias']
Traceback (most recent call last):
File "./transformers/examples/run_squad.py", line 614, in
main()
File "./transformers/examples/run_squad.py", line 532, in main
cache_dir=args.cache_dir if args.cache_dir else None)
File "/HOME/sandpit/transformers/transformers/modeling_utils.py", line 486, in from_pretrained
model.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for XLMForQuestionAnswering:
size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).
```

Source

waalge

All 14 comments

I've tried different XLM* models with and I've obtained the same error you've. I suspect it's broken something into the implementation of XLM* models or the .bin file uploaded to AWS S3.
N.B: I've tried to load xlm-mlm-17-1280 with the usual procedure (i.e. by using from_pretrained method) which works as expected in #2043 (about 15 days ago), but now it doesn't work (same error). Therefore, there's something broken for sure.
N.B: it's not a download problem itself, I've tried also with force_download=True parameter.

The stack trace is the following:

Python 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 19:07:31) 
[GCC 7.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import transformers
/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint8 = np.dtype([("qint8", np.int8, 1)])
/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint16 = np.dtype([("qint16", np.int16, 1)])
/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint32 = np.dtype([("qint32", np.int32, 1)])
/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  np_resource = np.dtype([("resource", np.ubyte, 1)])
>>> from transformers import XLMTokenizer, XLMWithLMHeadModel
>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-ende-1024')
Downloading: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.44M/1.44M [00:00<00:00, 2.06MB/s]
Downloading: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.00M/1.00M [00:00<00:00, 1.71MB/s]
>>> model = XLMWithLMHeadModel.from_pretrained('xlm-mlm-ende-1024')
Downloading: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 396/396 [00:00<00:00, 177kB/s]
Downloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 835M/835M [01:13<00:00, 11.3MB/s]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/<user>/Desktop/transformers/transformers/transformers/modeling_utils.py", line 486, in from_pretrained
    model.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for XLMWithLMHeadModel:
    size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).
    size mismatch for pred_layer.proj.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).
    size mismatch for pred_layer.proj.bias: copying a param with shape torch.Size([64699]) from checkpoint, the shape in current model is torch.Size([30145]).
>>> from transformers import XLMTokenizer, XLMForQuestionAnswering
>>> model = XLMForQuestionAnswering.from_pretrained('xlm-mlm-ende-1024')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/<user>/Desktop/transformers/transformers/transformers/modeling_utils.py", line 486, in from_pretrained
    model.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for XLMForQuestionAnswering:
    size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).
>>> model = XLMForQuestionAnswering.from_pretrained('xlm-clm-ende-1024')
Downloading: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 396/396 [00:00<00:00, 164kB/s]
Downloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 835M/835M [01:11<00:00, 11.7MB/s]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/<user>/Desktop/transformers/transformers/transformers/modeling_utils.py", line 486, in from_pretrained
    model.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for XLMForQuestionAnswering:
    size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).

Bug

Model: XLM_mlm_ende_1024

Language: English

Using: the official example scripts: run_squad.py

To Reproduce

Steps to reproduce the behavior:

install dependecies and download squad v1.1 data; pull, install transformers from github master.
run run_squad.py with the following args

--model_type xlm --model_name_or_path xlm-mlm-ende-1024 --do_train --do_eval --train_file ./squad_data/train-v1.1.json --predict_file ./squad_data/dev-v1.1.json --per_gpu_train_batch_size 12 --learning_rate 3e-5 --num_train_epochs 2.0 --max_seq_length 384 --doc_stride 128 --output_dir ./debug_xlm

Ultimate error:

size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).

Full inout ouput below.

Expected behavior: finetune xlm for squad.

Environment

OS: OpenSuse 15.0
Python version: 3.6
PyTorch version: torch.version = '1.3.1+cpu'
PyTorch Transformers version (or branch): (just transformers now?) 2.2.2
Using GPU ? nope
Distributed of parallel setup ? Ummm n/a

Relates to previous issues (possibly):

I've had with XLM
Similar looking error

Additional context

12/18/2019 15:47:23 - WARNING - __main__ -   Process rank: -1, device: cpu, n_gpu: 0, distributed training: False, 16-bits training: False
12/18/2019 15:47:23 - INFO - transformers.configuration_utils -   loading configuration file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-config.json from cache at /HOME/.cache/torch/transformers/8f689e7cdf34bbebea67ad44ad6a142c9c5144e5c19d989839139e0d47d1ed74.0038e5c2b48fc777632fc95c3d3422203693750b1d0845a511b3bb84ad6d8c29
12/18/2019 15:47:23 - INFO - transformers.configuration_utils -   Model config {
  "asm": false,
  "attention_dropout": 0.1,
  "bos_index": 0,
  "causal": false,
  "dropout": 0.1,
  "emb_dim": 1024,
  "embed_init_std": 0.02209708691207961,
  "end_n_top": 5,
  "eos_index": 1,
  "finetuning_task": null,
  "gelu_activation": true,
  "id2label": {
    "0": "LABEL_0",
    "1": "LABEL_1"
  },
  "id2lang": {
    "0": "de",
    "1": "en"
  },
  "init_std": 0.02,
  "is_decoder": false,
  "is_encoder": true,
  "label2id": {
    "LABEL_0": 0,
    "LABEL_1": 1
  },
  "lang2id": {
    "de": 0,
    "en": 1
  },
  "layer_norm_eps": 1e-12,
  "mask_index": 5,
  "max_position_embeddings": 512,
  "max_vocab": -1,
  "min_count": 0,
  "n_heads": 8,
  "n_langs": 2,
  "n_layers": 6,
  "num_labels": 2,
  "output_attentions": false,
  "output_hidden_states": false,
  "output_past": true,
  "pad_index": 2,
  "pruned_heads": {},
  "same_enc_dec": true,
  "share_inout_emb": true,
  "sinusoidal_embeddings": false,
  "start_n_top": 5,
  "summary_activation": null,
  "summary_first_dropout": 0.1,
  "summary_proj_to_labels": true,
  "summary_type": "first",
  "summary_use_proj": true,
  "torchscript": false,
  "unk_index": 3,
  "use_bfloat16": false,
  "use_lang_emb": true,
  "vocab_size": 30145
}

12/18/2019 15:47:24 - INFO - transformers.tokenization_utils -   loading file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-vocab.json from cache at /HOME/.cache/torch/transformers/6771b710c1daf9d51643260fdf576f6353369c3563bf0fb12176c692778dca3f.2c29a4b393decdd458e6a9744fa1d6b533212e4003a4012731d2bc2261dc35f3
12/18/2019 15:47:24 - INFO - transformers.tokenization_utils -   loading file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-merges.txt from cache at /HOME/.cache/torch/transformers/85d878ffb1bc2c3395b785d10ce7fc91452780316140d7a26201d7a912483e44.42fa32826c068642fdcf24adbf3ef8158b3b81e210a3d03f3102cf5a899f92a0
12/18/2019 15:47:25 - INFO - transformers.modeling_utils -   loading weights file https://s3.amazonaws.com/models.huggingface.co/bert/xlm-mlm-ende-1024-pytorch_model.bin from cache at /HOME/.cache/torch/transformers/ea4c0bbee310b490decb2b608a4dbc8ed9f2e4a103dd729ce183770b0fef698b.119d74257b953e5d50d73555a430ced11b1c149a7c17583219935ec1bd37d948
12/18/2019 15:47:28 - INFO - transformers.modeling_utils -   Weights of XLMForQuestionAnswering not initialized from pretrained model: ['qa_outputs.start_logits.dense.weight', 'qa_outputs.start_logits.dense.bias', 'qa_outputs.end_logits.dense_0.weight', 'qa_outputs.end_logits.dense_0.bias', 'qa_outputs.end_logits.LayerNorm.weight', 'qa_outputs.end_logits.LayerNorm.bias', 'qa_outputs.end_logits.dense_1.weight', 'qa_outputs.end_logits.dense_1.bias', 'qa_outputs.answer_class.dense_0.weight', 'qa_outputs.answer_class.dense_0.bias', 'qa_outputs.answer_class.dense_1.weight']
12/18/2019 15:47:28 - INFO - transformers.modeling_utils -   Weights from pretrained model not used in XLMForQuestionAnswering: ['pred_layer.proj.weight', 'pred_layer.proj.bias']
Traceback (most recent call last):
  File "./transformers/examples/run_squad.py", line 614, in <module>
    main()
  File "./transformers/examples/run_squad.py", line 532, in main
    cache_dir=args.cache_dir if args.cache_dir else None)
  File "/HOME/sandpit/transformers/transformers/modeling_utils.py", line 486, in from_pretrained
    model.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for XLMForQuestionAnswering:
        size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).

TheEdoardo93 on 18 Dec 2019

Indeed, there seems to be an error that was introduced by #2164. I'm looking into it now. Thanks for raising an issue!

LysandreJik on 18 Dec 2019

👍1

Please let me know if 8efc6dd fixes this issue!

LysandreJik on 18 Dec 2019

🎉1

I've installed Transformers from source (master branch) with pip install git+https://github.com/huggingface/transformers.git right now, but it seems to be the same bug. Is it possible? The stack trace is the same as before. @LysandreJik

Please let me know if 8efc6dd fixes this issue!

TheEdoardo93 on 18 Dec 2019

Hmm could you post a short snippet to reproduce? Running your initial script in my environment doesn't raise any error:

from transformers import XLMWithLMHeadModel
XLMWithLMHeadModel.from_pretrained("xlm-mlm-17-1280")

The error seems to be fixed on my side

LysandreJik on 18 Dec 2019

I'm trying to use XLMForQuestionAnswering model, is it right for run_squad.py correct?

>>> import transformers
>>> from transformers import XLMForQuestionAnswering
>>> model = XLMForQuestionAnswering.from_pretrained('xlm-mlm-ende-1024', force_download=True)
Downloading: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 396/396 [00:00<00:00, 146kB/s]
Downloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 835M/835M [01:16<00:00, 10.9MB/s]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/vidiemme/Desktop/transformers/transformers/transformers/modeling_utils.py", line 486, in from_pretrained
    model.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for XLMForQuestionAnswering:
    size mismatch for transformer.embeddings.weight: copying a param with shape torch.Size([64699, 1024]) from checkpoint, the shape in current model is torch.Size([30145, 1024]).

N.B: I've tried also your piece of code in my environment but it doesn't work (the same bug as before). How is it possible? I'm using Python 3.6.9, OS Ubuntu 16.04, PyTorch 1.3.1 and TensorFlow 2.0.

Hmm could you post a short snippet to reproduce? Running your initial script in my environment doesn't raise any error:
from transformers import XLMWithLMHeadModel
XLMWithLMHeadModel.from_pretrained("xlm-mlm-17-1280")
The error seems to be fixed on my side

TheEdoardo93 on 18 Dec 2019

Indeed, it doesn't fail on my side either. Is there any way you could go check in your environment, I guess (according to your error trace) following the path:

/home/vidiemme/Desktop/transformers/transformers/transformers/configuration_xlm.py

and telling me if the following lines:

        if "n_words" in kwargs:
            self.n_words = kwargs["n_words"]

Are on lines 147-148? Just to make sure the install from source worked correctly. Thank you @TheEdoardo93

LysandreJik on 18 Dec 2019

N.B: I've tried also your piece of code in my environment but it doesn't work (the same bug as before). How is it possible? I'm using Python 3.6.9, OS Ubuntu 16.04, PyTorch 1.3.1 and TensorFlow 2.0.

Hmm okay, I'm looking into it.

LysandreJik on 18 Dec 2019

In the file you've said to me at line 147-148 I've got the following lines:

@property
    def n_words(self):  # For backward compatibility
        return self.vocab_size

I don't have the lines you've posted above. Therefore, I can say that I haven't installed the Transformers library correctly. How can I do (i.e. install from master after your fix)? Usually I do the following: pip install git+https://github.com/huggingface/transformers.git

Indeed, it doesn't fail on my side either. Is there any way you could go check in your environment, I guess (according to your error trace) following the path:
/home/vidiemme/Desktop/transformers/transformers/transformers/configuration_xlm.py
and telling me if the following lines:
        if "n_words" in kwargs:
            self.n_words = kwargs["n_words"]
Are on lines 147-148? Just to make sure the install from source worked correctly. Thank you @TheEdoardo93

TheEdoardo93 on 18 Dec 2019

Hmm it seems your install from source didn't work. I don't exactly know how your environment is setup, but it looks like you've cloned the repository and the code is running from this clone rather than from the library installed in your environment/virtual environment.

If you did clone it in /home/vidiemme/Desktop/transformers/, I would just do a git pull to update it.

LysandreJik on 18 Dec 2019

Now it works as expected! Your fix fixes the bug! Great work! You can close this issue for me ;)

Now we can import both XLMForQuestionAnswering.from_pretrained('xlm-mlm-ende-1024') and XLMWithLMHeadModel.from_pretrained("xlm-mlm-17-1280") correctly.

Hmm it seems your install from source didn't work. I don't exactly know how your environment is setup, but it looks like you've cloned the repository and the code is running from this clone rather than from the library installed in your environment/virtual environment.

If you did clone it in /home/vidiemme/Desktop/transformers/, I would just do a git pull to update it.

TheEdoardo93 on 18 Dec 2019

Glad to hear that @TheEdoardo93 !

LysandreJik on 18 Dec 2019

👍1

A bit late to the party, but I can provide a second confirmation that this error no longer appears.
Thanks!

waalge on 19 Dec 2019

PS I don't know where is a useful place to put this but for anyone training XLM on squad....

The command above now runs to completion.
Its score is underwhelming but demonstrates some training has been achieved

Results: {'exact': 56.9441816461684, 'f1': 67.90690126118979, 'total': 10570, 'HasAns_exact': 56.9441816461684, 'HasAns_f1': 67.90690126118979, 'HasAns_total': 10570, 'best_exact': 56.9441816461684, 'best_exact_thresh': 0.0, 'best_f1': 67.90690126118979, 'best_f1_thresh': 0.0}

waalge on 20 Dec 2019

Was this page helpful?

0 / 5 - 0 ratings