Haystack: How to use XLM-R as retriever correctly?
I'm trying to use xlm-r-100langs-bert-base-nli-stsb-mean-tokens as retriever with
retriever = EmbeddingRetriever(document_store=document_store, embedding_model='xlm-r-100langs-bert-base-nli-stsb-mean-tokens', model_format='sentence_transformers')
when I try to embed a text with retriever.embed('test'), it raises this error:
/usr/local/lib/python3.6/dist-packages/transformers/modeling_utils.py in get_extended_attention_mask(self, attention_mask, input_shape, device)
260 raise ValueError(
261 "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
--> 262 input_shape, attention_mask.shape
263 )
264 )
ValueError: Wrong shape for input_ids (shape torch.Size([4])) or attention_mask (shape torch.Size([4]))
I also tried to use the model from huggingface model hub:
retriever = EmbeddingRetriever(document_store=document_store, embedding_model='sentence-transformers/xlm-r-100langs-bert-base-nli-stsb-mean-tokens', model_format='transformers')
but it raises this error:
TypeError Traceback (most recent call last)
<ipython-input-34-0b021b13e848> in <module>()
1 from haystack.retriever.dense import EmbeddingRetriever
----> 2 retriever = EmbeddingRetriever(document_store=document_store, embedding_model='sentence-transformers/xlm-r-100langs-bert-base-nli-stsb-mean-tokens', model_format='transformers')
6 frames
/usr/local/lib/python3.6/dist-packages/haystack/retriever/dense.py in __init__(self, document_store, embedding_model, use_gpu, model_format, pooling_strategy, emb_extraction_layer)
300 self.embedding_model = Inferencer.load(
301 embedding_model, task_type="embeddings", extraction_strategy=self.pooling_strategy,
--> 302 extraction_layer=self.emb_extraction_layer, gpu=use_gpu, batch_size=4, max_seq_len=512, num_processes=0
303 )
304
/usr/local/lib/python3.6/dist-packages/farm/infer.py in load(cls, model_name_or_path, batch_size, gpu, task_type, return_class_probs, strict, max_seq_len, doc_stride, extraction_layer, extraction_strategy, s3e_stats, num_processes, disable_tqdm, tokenizer_class, use_fast, tokenizer_args, dummy_ph, benchmarking)
271 tokenizer_class=tokenizer_class,
272 use_fast=use_fast,
--> 273 **tokenizer_args,
274 )
275
/usr/local/lib/python3.6/dist-packages/farm/modeling/tokenization.py in load(cls, pretrained_model_name_or_path, tokenizer_class, use_fast, **kwargs)
131 ret = BertTokenizerFast.from_pretrained(pretrained_model_name_or_path, **kwargs)
132 else:
--> 133 ret = BertTokenizer.from_pretrained(pretrained_model_name_or_path, **kwargs)
134 elif tokenizer_class == "XLNetTokenizer":
135 if use_fast:
/usr/local/lib/python3.6/dist-packages/transformers/tokenization_utils_base.py in from_pretrained(cls, *inputs, **kwargs)
1423
1424 """
-> 1425 return cls._from_pretrained(*inputs, **kwargs)
1426
1427 @classmethod
/usr/local/lib/python3.6/dist-packages/transformers/tokenization_utils_base.py in _from_pretrained(cls, pretrained_model_name_or_path, *init_inputs, **kwargs)
1570 # Instantiate tokenizer.
1571 try:
-> 1572 tokenizer = cls(*init_inputs, **init_kwargs)
1573 except OSError:
1574 raise OSError(
/usr/local/lib/python3.6/dist-packages/transformers/tokenization_bert.py in __init__(self, vocab_file, do_lower_case, do_basic_tokenize, never_split, unk_token, sep_token, pad_token, cls_token, mask_token, tokenize_chinese_chars, strip_accents, **kwargs)
189 )
190
--> 191 if not os.path.isfile(vocab_file):
192 raise ValueError(
193 "Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained "
/usr/lib/python3.6/genericpath.py in isfile(path)
28 """Test whether a path is a regular file"""
29 try:
---> 30 st = os.stat(path)
31 except OSError:
32 return False
TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType
Any advice how to use the xlm-r-100langs-bert-base-nli-stsb-mean-tokens model correctly?
bug
khalidbhs
All 7 comments
Thanks for reporting this bug @khalidbhs !
For
model_format='transformers': Edge case where the model name contains "bert" and therefore the BERT tokenizer is loaded instead of the XLMR tokenizer. We'll fix this.For
model_format='sentence-transformers': Not clear yet what is happening here. Possibly some version issues. What transformers + sentence-transformers version do you use?
tholor
on 20 Oct 2020
❤1
Cool, thanks!
I used transformers==3.1.0 and sentence-transformers==0.3.4
Here's the full stack trace for model_format='sentence-transformers' btw:
ValueError Traceback (most recent call last)
<ipython-input-24-d3ac4f6b76bb> in <module>()
----> 1 retriever.embed('test')
7 frames
/usr/local/lib/python3.6/dist-packages/haystack/retriever/dense.py in embed(self, texts)
347 # text is single string, sentence-transformers needs a list of strings
348 # get back list of numpy embedding vectors
--> 349 emb = self.embedding_model.encode(texts) # type: ignore
350 emb = [r for r in emb]
351 return emb
/usr/local/lib/python3.6/dist-packages/sentence_transformers/SentenceTransformer.py in encode(self, sentences, batch_size, show_progress_bar, output_value, convert_to_numpy, convert_to_tensor, is_pretokenized, device, num_workers)
150
151 with torch.no_grad():
--> 152 out_features = self.forward(features)
153 embeddings = out_features[output_value]
154
/usr/local/lib/python3.6/dist-packages/torch/nn/modules/container.py in forward(self, input)
115 def forward(self, input):
116 for module in self:
--> 117 input = module(input)
118 return input
119
/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
720 result = self._slow_forward(*input, **kwargs)
721 else:
--> 722 result = self.forward(*input, **kwargs)
723 for hook in itertools.chain(
724 _global_forward_hooks.values(),
/usr/local/lib/python3.6/dist-packages/sentence_transformers/models/Transformer.py in forward(self, features)
33 def forward(self, features):
34 """Returns token_embeddings, cls_token"""
---> 35 output_states = self.auto_model(**features)
36 output_tokens = output_states[0]
37
/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
720 result = self._slow_forward(*input, **kwargs)
721 else:
--> 722 result = self.forward(*input, **kwargs)
723 for hook in itertools.chain(
724 _global_forward_hooks.values(),
/usr/local/lib/python3.6/dist-packages/transformers/modeling_bert.py in forward(self, input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, encoder_hidden_states, encoder_attention_mask, output_attentions, output_hidden_states, return_dict)
802 # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
803 # ourselves in which case we just need to make it broadcastable to all heads.
--> 804 extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
805
806 # If a 2D ou 3D attention mask is provided for the cross-attention
/usr/local/lib/python3.6/dist-packages/transformers/modeling_utils.py in get_extended_attention_mask(self, attention_mask, input_shape, device)
260 raise ValueError(
261 "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
--> 262 input_shape, attention_mask.shape
263 )
264 )
ValueError: Wrong shape for input_ids (shape torch.Size([4])) or attention_mask (shape torch.Size([4]))
Fixing this in https://github.com/deepset-ai/FARM/issues/571
tholor
on 21 Oct 2020
👍1
bogdankostic
on 28 Oct 2020
Will be available in Haystack within the next days (after the FARM release) or you install the latest FARM version from master manually.
tholor
on 28 Oct 2020
Great, I've tried it and it's working now, thanks!
khalidbhs
on 31 Oct 2020
Perfect, thanks for the feedback!
tholor
on 31 Oct 2020
Was this page helpful?
0 / 5 - 0 ratings
Related issues
rshtirmer
路
3Comments
tobiasbj
路
6Comments
dany-nonstop
路
7Comments
tholor
路
3Comments
antoniolanza1996
路
4Comments