Buckets:
| import"../chunks/DsnmJJEf.js";import{i as Qe,h as Se,C as Pe,H as d,D as t,E as Ye,s as Ae,a as p}from"../chunks/D6eQvJIN.js";import{p as He,o as Ee,s as e,f as c,a as l,b as Le,c as s,d as f,n as g,r as a}from"../chunks/D9TCMYxZ.js";import{T as u}from"../chunks/eoUhp3Ag.js";import{E as m}from"../chunks/D8azpo8f.js";const Oe='{"title":"Funnel Transformer","local":"funnel-transformer","sections":[{"title":"Overview","local":"overview","sections":[],"depth":2},{"title":"Usage tips","local":"usage-tips","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2},{"title":"FunnelConfig","local":"transformers.FunnelConfig","sections":[],"depth":2},{"title":"FunnelTokenizer","local":"transformers.FunnelTokenizer","sections":[],"depth":2},{"title":"Funnel specific outputs","local":"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput","sections":[],"depth":2},{"title":"FunnelBaseModel","local":"transformers.FunnelBaseModel","sections":[],"depth":2},{"title":"FunnelModel","local":"transformers.FunnelModel","sections":[],"depth":2},{"title":"FunnelModelForPreTraining","local":"transformers.FunnelForPreTraining","sections":[],"depth":2},{"title":"FunnelForMaskedLM","local":"transformers.FunnelForMaskedLM","sections":[],"depth":2},{"title":"FunnelForSequenceClassification","local":"transformers.FunnelForSequenceClassification","sections":[],"depth":2},{"title":"FunnelForMultipleChoice","local":"transformers.FunnelForMultipleChoice","sections":[],"depth":2},{"title":"FunnelForTokenClassification","local":"transformers.FunnelForTokenClassification","sections":[],"depth":2},{"title":"FunnelForQuestionAnswering","local":"transformers.FunnelForQuestionAnswering","sections":[],"depth":2}],"depth":1}';var De=f('<meta name="hf:doc:metadata"/>'),h=f(`<p>Although the recipe for forward pass needs to be defined within this function, one should call the <code>Module</code> instance afterwards instead of this since the former takes care of running the pre and post processing steps while | |
| the latter silently ignores them.</p>`),_=f("<p>Example:</p> <!>",1),Ke=f("<p>Examples:</p> <!>",1),$e=f("<p>Example of single-label classification:</p> <!>",1),en=f("<p>Example of multi-label classification:</p> <!>",1),nn=f(`<p></p> <p><em>This model was published in HF papers on 2020-06-05 and contributed to Hugging Face Transformers on 2020-11-16.</em></p> <!> <!> <!> <p>The Funnel Transformer model was proposed in the paper <a href="https://huggingface.co/papers/2006.03236" rel="nofollow">Funnel-Transformer: Filtering out Sequential Redundancy for | |
| Efficient Language Processing</a>. It is a bidirectional transformer model, like | |
| BERT, but with a pooling operation after each block of layers, a bit like in traditional convolutional neural networks | |
| (CNN) in computer vision.</p> <p>The abstract from the paper is the following:</p> <p><em>With the success of language pretraining, it is highly desirable to develop more efficient architectures of good | |
| scalability that can exploit the abundant unlabeled data at a lower cost. To improve the efficiency, we examine the | |
| much-overlooked redundancy in maintaining a full-length token-level presentation, especially for tasks that only | |
| require a single-vector presentation of the sequence. With this intuition, we propose Funnel-Transformer which | |
| gradually compresses the sequence of hidden states to a shorter one and hence reduces the computation cost. More | |
| importantly, by re-investing the saved FLOPs from length reduction in constructing a deeper or wider model, we further | |
| improve the model capacity. In addition, to perform token-level predictions as required by common pretraining | |
| objectives, Funnel-Transformer is able to recover a deep representation for each token from the reduced hidden sequence | |
| via a decoder. Empirically, with comparable or fewer FLOPs, Funnel-Transformer outperforms the standard Transformer on | |
| a wide variety of sequence-level prediction tasks, including text classification, language understanding, and reading | |
| comprehension.</em></p> <p>This model was contributed by <a href="https://huggingface.co/sgugger" rel="nofollow">sgugger</a>. The original code can be found <a href="https://github.com/laiguokun/Funnel-Transformer" rel="nofollow">here</a>.</p> <!> <ul><li>Since Funnel Transformer uses pooling, the sequence length of the hidden states changes after each block of layers. This way, their length is divided by 2, which speeds up the computation of the next hidden states. | |
| The base model therefore has a final sequence length that is a quarter of the original one. This model can be used | |
| directly for tasks that just require a sentence summary (like sequence classification or multiple choice). For other | |
| tasks, the full model is used; this full model has a decoder that upsamples the final hidden states to the same | |
| sequence length as the input.</li> <li>For tasks such as classification, this is not a problem, but for tasks like masked language modeling or token classification, we need a hidden state with the same sequence length as the original input. In those cases, the final hidden states are upsampled to the input sequence length and go through two additional layers. That’s why there are two versions of each checkpoint. The version suffixed with “-base” contains only the three blocks, while the version without that suffix contains the three blocks and the upsampling head with its additional layers.</li> <li>The Funnel Transformer checkpoints are all available with a full version and a base version. The first ones should be | |
| used for <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelModel">FunnelModel</a>, <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForPreTraining">FunnelForPreTraining</a>, <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForMaskedLM">FunnelForMaskedLM</a>, <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForTokenClassification">FunnelForTokenClassification</a> and <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForQuestionAnswering">FunnelForQuestionAnswering</a>. The second ones should be used for <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelBaseModel">FunnelBaseModel</a>, <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForSequenceClassification">FunnelForSequenceClassification</a> and <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForMultipleChoice">FunnelForMultipleChoice</a>.</li></ul> <!> <ul><li><a href="../tasks/sequence_classification">Text classification task guide</a></li> <li><a href="../tasks/token_classification">Token classification task guide</a></li> <li><a href="../tasks/question_answering">Question answering task guide</a></li> <li><a href="../tasks/masked_language_modeling">Masked language modeling task guide</a></li> <li><a href="../tasks/multiple_choice">Multiple choice task guide</a></li></ul> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>This is the configuration class to store the configuration of a (‘FunnelModel’, ‘FunnelBaseModel’). It is used to instantiate a Funnel | |
| model according to the specified arguments, defining the model architecture. Instantiating a configuration with the | |
| defaults will yield a similar configuration to that of the <a href="https://huggingface.co/funnel-transformer/small" rel="nofollow">funnel-transformer/small</a></p> <p>Configuration objects inherit from <a href="/docs/transformers/pr_36708/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> and can be used to control the model outputs. Read the | |
| documentation from <a href="/docs/transformers/pr_36708/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> for more information.</p></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Construct a Funnel Transformer tokenizer (backed by HuggingFace’s tokenizers library). Based on WordPiece.</p> <p>This tokenizer inherits from <a href="/docs/transformers/pr_36708/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a> which contains most of the main methods. Users should | |
| refer to this superclass for more information regarding those methods.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Retrieve sequence ids from a token list that has no special tokens added.</p> <p>For fast tokenizers, data collators call this with <code>already_has_special_tokens=True</code> to build a mask over an | |
| already-formatted sequence. In that case, we compute the mask by checking membership in <code>all_special_ids</code>.</p></div> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Output type of <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForPreTraining">FunnelForPreTraining</a>.</p></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The base Funnel Transformer Model transformer outputting raw hidden-states without upsampling head (also called | |
| decoder) or any task-specific head on top.</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelBaseModel">FunnelBaseModel</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>last_hidden_state</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>) — Sequence of hidden-states at the output of the last layer of the model.</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The bare Funnel Model outputting raw hidden-states without any specific head on top.</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelModel">FunnelModel</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>last_hidden_state</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>) — Sequence of hidden-states at the output of the last layer of the model.</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Funnel Transformer model with a binary classification head on top as used during pretraining for identifying | |
| generated tokens.</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForPreTraining">FunnelForPreTraining</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>*optional*</code>, returned when <code>labels</code> is provided, <code>torch.FloatTensor</code> of shape <code>(1,)</code>) — Total loss of the ELECTRA-style objective.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Prediction scores of the head (scores for each token before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple[torch.FloatTensor]</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple[torch.FloatTensor]</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Funnel Model with a <code>language modeling</code> head on top.”</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForMaskedLM">FunnelForMaskedLM</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Masked language modeling (MLM) loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.vocab_size)</code>) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Funnel Transformer Model with a sequence classification/regression head on top (two linear layer on top of the | |
| first timestep of the last hidden state) e.g. for GLUE tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForSequenceClassification">FunnelForSequenceClassification</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification (or regression if config.num_labels==1) loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, config.num_labels)</code>) — Classification (or regression if config.num_labels==1) scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Funnel Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a | |
| softmax) e.g. for RocStories/SWAG tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForMultipleChoice">FunnelForMultipleChoice</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <em>(1,)</em>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_choices)</code>) — <em>num_choices</em> is the second dimension of the input tensors. (see <em>input_ids</em> above).</p> <p>Classification scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Funnel transformer with a token classification head on top (a linear layer on top of the hidden-states | |
| output) e.g. for Named-Entity-Recognition (NER) tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForTokenClassification">FunnelForTokenClassification</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.num_labels)</code>) — Classification scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Funnel transformer with a span classification head on top for extractive question-answering tasks like | |
| SQuAD (a linear layer on top of the hidden-states output to compute <code>span start logits</code> and <code>span end logits</code>).</p> <p>This model inherits from <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the | |
| library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads | |
| etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass. | |
| Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage | |
| and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelForQuestionAnswering">FunnelForQuestionAnswering</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.</p></li> <li><p><strong>start_logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-start scores (before SoftMax).</p></li> <li><p><strong>end_logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-end scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.</p></li></ul> <!></div></div> <!> <p></p>`,1);function ln(ze,Je){He(Je,!1),Ee(()=>{new URLSearchParams(window.location.search).get("fw")}),Qe();var C=nn();Se("1a9aaqq",o=>{var r=De();Ae(r,"content",Oe),l(o,r)});var W=e(c(C),4);Pe(W,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var x=e(W,2);d(x,{title:"Funnel Transformer",local:"funnel-transformer",headingTag:"h1"});var Z=e(x,2);d(Z,{title:"Overview",local:"overview",headingTag:"h2"});var B=e(Z,10);d(B,{title:"Usage tips",local:"usage-tips",headingTag:"h2"});var q=e(B,4);d(q,{title:"Resources",local:"resources",headingTag:"h2"});var N=e(q,4);d(N,{title:"FunnelConfig",local:"transformers.FunnelConfig",headingTag:"h2"});var b=e(N,2),Ue=s(b);t(Ue,{name:"class transformers.FunnelConfig",anchor:"transformers.FunnelConfig",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/configuration_funnel.py#L24",parameters:[{name:"transformers_version",val:": str | None = None"},{name:"architectures",val:": list[str] | None = None"},{name:"output_hidden_states",val:": bool | None = False"},{name:"return_dict",val:": bool | None = True"},{name:"dtype",val:": str | torch.dtype | None = None"},{name:"chunk_size_feed_forward",val:": int = 0"},{name:"is_encoder_decoder",val:": bool = False"},{name:"id2label",val:": dict[int, str] | dict[str, str] | None = None"},{name:"label2id",val:": dict[str, int] | dict[str, str] | None = None"},{name:"problem_type",val:": Literal['regression', 'single_label_classification', 'multi_label_classification'] | None = None"},{name:"vocab_size",val:": int = 30522"},{name:"block_sizes",val:": list[int] | tuple[int, ...] = (4, 4, 4)"},{name:"block_repeats",val:": list[int] | None = None"},{name:"num_decoder_layers",val:": int = 2"},{name:"d_model",val:": int = 768"},{name:"n_head",val:": int = 12"},{name:"d_head",val:": int = 64"},{name:"d_inner",val:": int = 3072"},{name:"hidden_act",val:": str = 'gelu_new'"},{name:"hidden_dropout",val:": float | int = 0.1"},{name:"attention_dropout",val:": float | int = 0.1"},{name:"activation_dropout",val:": float | int = 0.0"},{name:"initializer_range",val:": float = 0.1"},{name:"initializer_std",val:": float | None = None"},{name:"layer_norm_eps",val:": float = 1e-09"},{name:"pooling_type",val:": str = 'mean'"},{name:"attention_type",val:": str = 'relative_shift'"},{name:"separate_cls",val:": bool = True"},{name:"truncate_seq",val:": bool = True"},{name:"pool_q_only",val:": bool = True"},{name:"pad_token_id",val:": int | None = None"},{name:"tie_word_embeddings",val:": bool = True"}],parametersDescription:[{anchor:"transformers.FunnelConfig.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>30522</code>) — | |
| Vocabulary size of the model. Defines the number of different tokens that can be represented by the <code>input_ids</code>.`,name:"vocab_size"},{anchor:"transformers.FunnelConfig.block_sizes",description:`<strong>block_sizes</strong> (<code>list[int]</code>, <em>optional</em>, defaults to <code>[4, 4, 4]</code>) — | |
| The sizes of the blocks used in the model.`,name:"block_sizes"},{anchor:"transformers.FunnelConfig.block_repeats",description:`<strong>block_repeats</strong> (<code>list[int]</code>, <em>optional</em>) — | |
| If passed along, each layer of each block is repeated the number of times indicated.`,name:"block_repeats"},{anchor:"transformers.FunnelConfig.num_decoder_layers",description:`<strong>num_decoder_layers</strong> (<code>int</code>, <em>optional</em>, defaults to 2) — | |
| The number of layers in the decoder (when not using the base model).`,name:"num_decoder_layers"},{anchor:"transformers.FunnelConfig.d_model",description:`<strong>d_model</strong> (<code>int</code>, <em>optional</em>, defaults to <code>768</code>) — | |
| Size of the encoder layers and the pooler layer.`,name:"d_model"},{anchor:"transformers.FunnelConfig.n_head",description:`<strong>n_head</strong> (<code>int</code>, <em>optional</em>, defaults to <code>12</code>) — | |
| Number of attention heads for each attention layer in the Transformer decoder.`,name:"n_head"},{anchor:"transformers.FunnelConfig.d_head",description:`<strong>d_head</strong> (<code>int</code>, <em>optional</em>, defaults to <code>64</code>) — | |
| The attention head dimension. If None, it will default to hidden_size // num_attention_heads`,name:"d_head"},{anchor:"transformers.FunnelConfig.d_inner",description:`<strong>d_inner</strong> (<code>int</code>, <em>optional</em>, defaults to <code>3072</code>) — | |
| Dimension of the MLP representations.`,name:"d_inner"},{anchor:"transformers.FunnelConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code>, <em>optional</em>, defaults to <code>gelu_new</code>) — | |
| The non-linear activation function (function or string) in the decoder. For example, <code>"gelu"</code>, | |
| <code>"relu"</code>, <code>"silu"</code>, etc.`,name:"hidden_act"},{anchor:"transformers.FunnelConfig.hidden_dropout",description:`<strong>hidden_dropout</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.1</code>) — | |
| The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.`,name:"hidden_dropout"},{anchor:"transformers.FunnelConfig.attention_dropout",description:`<strong>attention_dropout</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.1</code>) — | |
| The dropout ratio for the attention probabilities.`,name:"attention_dropout"},{anchor:"transformers.FunnelConfig.activation_dropout",description:`<strong>activation_dropout</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.0</code>) — | |
| The dropout ratio for activations inside the fully connected layer.`,name:"activation_dropout"},{anchor:"transformers.FunnelConfig.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to <code>0.1</code>) — | |
| The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.FunnelConfig.initializer_std",description:`<strong>initializer_std</strong> (<code>float</code>, <em>optional</em>) — | |
| The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_std"},{anchor:"transformers.FunnelConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to <code>1e-09</code>) — | |
| The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.FunnelConfig.pooling_type",description:`<strong>pooling_type</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"mean"</code>) — | |
| Possible values are <code>"mean"</code> or <code>"max"</code>. The way pooling is performed at the beginning of each block.`,name:"pooling_type"},{anchor:"transformers.FunnelConfig.attention_type",description:`<strong>attention_type</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"relative_shift"</code>) — | |
| Possible values are <code>"relative_shift"</code> or <code>"factorized"</code>. The former is faster on CPU/GPU while the latter | |
| is faster on TPU.`,name:"attention_type"},{anchor:"transformers.FunnelConfig.separate_cls",description:`<strong>separate_cls</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether or not to separate the cls token when applying pooling.`,name:"separate_cls"},{anchor:"transformers.FunnelConfig.truncate_seq",description:`<strong>truncate_seq</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| When using <code>separate_cls</code>, whether or not to truncate the last token when pooling, to avoid getting a | |
| sequence length that is not a multiple of 2.`,name:"truncate_seq"},{anchor:"transformers.FunnelConfig.pool_q_only",description:`<strong>pool_q_only</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether or not to apply the pooling only to the query or to query, key and values for the attention layers.`,name:"pool_q_only"},{anchor:"transformers.FunnelConfig.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>) — | |
| Token id used for padding in the vocabulary.`,name:"pad_token_id"},{anchor:"transformers.FunnelConfig.tie_word_embeddings",description:`<strong>tie_word_embeddings</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether to tie weight embeddings according to model’s <code>tied_weights_keys</code> mapping.`,name:"tie_word_embeddings"}]}),g(4),a(b);var I=e(b,2);d(I,{title:"FunnelTokenizer",local:"transformers.FunnelTokenizer",headingTag:"h2"});var y=e(I,2),G=s(y);t(G,{name:"class transformers.FunnelTokenizer",anchor:"transformers.FunnelTokenizer",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/tokenization_funnel.py#L41",parameters:[{name:"vocab",val:": str | dict[str, int] | None = None"},{name:"do_lower_case",val:": bool = True"},{name:"unk_token",val:": str = '<unk>'"},{name:"sep_token",val:": str = '<sep>'"},{name:"pad_token",val:": str = '<pad>'"},{name:"cls_token",val:": str = '<cls>'"},{name:"mask_token",val:": str = '<mask>'"},{name:"bos_token",val:": str = '<s>'"},{name:"eos_token",val:": str = '</s>'"},{name:"clean_text",val:": bool = True"},{name:"tokenize_chinese_chars",val:": bool = True"},{name:"strip_accents",val:": bool | None = None"},{name:"wordpieces_prefix",val:": str = '##'"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelTokenizer.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>) — | |
| File containing the vocabulary.`,name:"vocab_file"},{anchor:"transformers.FunnelTokenizer.do_lower_case",description:`<strong>do_lower_case</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether or not to lowercase the input when tokenizing.`,name:"do_lower_case"},{anchor:"transformers.FunnelTokenizer.unk_token",description:`<strong>unk_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"<unk>"</code>) — | |
| The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this | |
| token instead.`,name:"unk_token"},{anchor:"transformers.FunnelTokenizer.sep_token",description:`<strong>sep_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"<sep>"</code>) — | |
| The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for | |
| sequence classification or for a text and a question for question answering. It is also used as the last | |
| token of a sequence built with special tokens.`,name:"sep_token"},{anchor:"transformers.FunnelTokenizer.pad_token",description:`<strong>pad_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"<pad>"</code>) — | |
| The token used for padding, for example when batching sequences of different lengths.`,name:"pad_token"},{anchor:"transformers.FunnelTokenizer.cls_token",description:`<strong>cls_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"<cls>"</code>) — | |
| The classifier token which is used when doing sequence classification (classification of the whole sequence | |
| instead of per-token classification). It is the first token of the sequence when built with special tokens.`,name:"cls_token"},{anchor:"transformers.FunnelTokenizer.mask_token",description:`<strong>mask_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"<mask>"</code>) — | |
| The token used for masking values. This is the token used when training this model with masked language | |
| modeling. This is the token which the model will try to predict.`,name:"mask_token"},{anchor:"transformers.FunnelTokenizer.clean_text",description:`<strong>clean_text</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether or not to clean the text before tokenization by removing any control characters and replacing all | |
| whitespaces by the classic one.`,name:"clean_text"},{anchor:"transformers.FunnelTokenizer.tokenize_chinese_chars",description:`<strong>tokenize_chinese_chars</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) — | |
| Whether or not to tokenize Chinese characters. This should likely be deactivated for Japanese (see <a href="https://github.com/huggingface/transformers/issues/328" rel="nofollow">this | |
| issue</a>).`,name:"tokenize_chinese_chars"},{anchor:"transformers.FunnelTokenizer.bos_token",description:`<strong>bos_token</strong> (<code>str</code>, <code>optional</code>, defaults to <code>"<s>"</code>) — | |
| The beginning of sentence token.`,name:"bos_token"},{anchor:"transformers.FunnelTokenizer.eos_token",description:`<strong>eos_token</strong> (<code>str</code>, <code>optional</code>, defaults to <code>"</s>"</code>) — | |
| The end of sentence token.`,name:"eos_token"},{anchor:"transformers.FunnelTokenizer.strip_accents",description:`<strong>strip_accents</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to strip all accents. If this option is not specified, then it will be determined by the | |
| value for <code>lowercase</code> (as in the original BERT).`,name:"strip_accents"},{anchor:"transformers.FunnelTokenizer.wordpieces_prefix",description:`<strong>wordpieces_prefix</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"##"</code>) — | |
| The prefix for subwords.`,name:"wordpieces_prefix"},{anchor:"transformers.FunnelTokenizer.vocab",description:`<strong>vocab</strong> (<code>str</code> or <code>dict[str, int]</code>, <em>optional</em>) — | |
| Custom vocabulary dictionary.`,name:"vocab"}]});var T=e(G,6),Ce=s(T);t(Ce,{name:"get_special_tokens_mask",anchor:"transformers.FunnelTokenizer.get_special_tokens_mask",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/tokenization_utils_base.py#L1320",parameters:[{name:"token_ids_0",val:": list[int]"},{name:"token_ids_1",val:": list[int] | None = None"},{name:"already_has_special_tokens",val:": bool = False"}],parametersDescription:[{anchor:"transformers.FunnelTokenizer.get_special_tokens_mask.token_ids_0",description:"<strong>token_ids_0</strong> — List of IDs for the (possibly already formatted) sequence.",name:"token_ids_0"},{anchor:"transformers.FunnelTokenizer.get_special_tokens_mask.token_ids_1",description:"<strong>token_ids_1</strong> — Unused when <code>already_has_special_tokens=True</code>. Must be None in that case.",name:"token_ids_1"},{anchor:"transformers.FunnelTokenizer.get_special_tokens_mask.already_has_special_tokens",description:"<strong>already_has_special_tokens</strong> — Whether the sequence is already formatted with special tokens.",name:"already_has_special_tokens"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>1 for a special token, 0 for a sequence token.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A list of integers in the range [0, 1]</p> | |
| `}),g(4),a(T);var R=e(T,2),We=s(R);t(We,{name:"save_vocabulary",anchor:"transformers.FunnelTokenizer.save_vocabulary",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/tokenization_utils_tokenizers.py#L545",parameters:[{name:"save_directory",val:": str"},{name:"filename_prefix",val:": str | None = None"}]}),a(R),a(y);var X=e(y,2);d(X,{title:"Funnel specific outputs",local:"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput",headingTag:"h2"});var M=e(X,2),xe=s(M);t(xe,{name:"class transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput",anchor:"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L727",parameters:[{name:"loss",val:": typing.Optional[torch.FloatTensor] = None"},{name:"logits",val:": typing.Optional[torch.FloatTensor] = None"},{name:"hidden_states",val:": tuple[torch.FloatTensor] | None = None"},{name:"attentions",val:": tuple[torch.FloatTensor] | None = None"}],parametersDescription:[{anchor:"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput.loss",description:`<strong>loss</strong> (<code>*optional*</code>, returned when <code>labels</code> is provided, <code>torch.FloatTensor</code> of shape <code>(1,)</code>) — | |
| Total loss of the ELECTRA-style objective.`,name:"loss"},{anchor:"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput.logits",description:`<strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — | |
| Prediction scores of the head (scores for each token before SoftMax).`,name:"logits"},{anchor:"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput.hidden_states",description:`<strong>hidden_states</strong> (<code>tuple[torch.FloatTensor]</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — | |
| Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, + | |
| one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> | |
| <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.`,name:"hidden_states"},{anchor:"transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput.attentions",description:`<strong>attentions</strong> (<code>tuple[torch.FloatTensor]</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — | |
| Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> | |
| <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention | |
| heads.`,name:"attentions"}]}),g(2),a(M);var V=e(M,2);d(V,{title:"FunnelBaseModel",local:"transformers.FunnelBaseModel",headingTag:"h2"});var k=e(V,2),Q=s(k);t(Q,{name:"class transformers.FunnelBaseModel",anchor:"transformers.FunnelBaseModel",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L747",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelBaseModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var S=e(Q,8),P=s(S);t(P,{name:"forward",anchor:"transformers.FunnelBaseModel.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L763",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelBaseModel.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelBaseModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelBaseModel.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelBaseModel.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p> | |
| <p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.FunnelBaseModel.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelBaseModel.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelBaseModel.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelBaseModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput" | |
| >BaseModelOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput" | |
| >BaseModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var Y=e(P,4);u(Y,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var Ze=e(Y,4);m(Ze,{anchor:"transformers.FunnelBaseModel.forward.example",children:(o,r)=>{var n=_(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelBaseModel | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelBaseModel.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"Hello, my dog is cute"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span>outputs = model(**inputs) | |
| <span class="hljs-meta">>>> </span>last_hidden_states = outputs.last_hidden_state`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(S),a(k);var A=e(k,2);d(A,{title:"FunnelModel",local:"transformers.FunnelModel",headingTag:"h2"});var w=e(A,2),H=s(w);t(H,{name:"class transformers.FunnelModel",anchor:"transformers.FunnelModel",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L814",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var E=e(H,8),L=s(E);t(L,{name:"forward",anchor:"transformers.FunnelModel.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L831",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelModel.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelModel.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelModel.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelModel.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelModel.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput" | |
| >BaseModelOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput" | |
| >BaseModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var Be=e(L,4);u(Be,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}}),g(2),a(E),a(w);var O=e(w,2);d(O,{title:"FunnelModelForPreTraining",local:"transformers.FunnelForPreTraining",headingTag:"h2"});var F=e(O,2),D=s(F);t(D,{name:"class transformers.FunnelForPreTraining",anchor:"transformers.FunnelForPreTraining",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L913",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelForPreTraining.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var K=e(D,8),$=s(K);t($,{name:"forward",anchor:"transformers.FunnelForPreTraining.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L922",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelForPreTraining.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelForPreTraining.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelForPreTraining.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelForPreTraining.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelForPreTraining.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Labels for computing the ELECTRA-style loss. Input should be a sequence of tokens (see <code>input_ids</code> | |
| docstring) Indices should be in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 indicates the token is an original token,</li> | |
| <li>1 indicates the token was replaced.</li> | |
| </ul>`,name:"labels"},{anchor:"transformers.FunnelForPreTraining.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelForPreTraining.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelForPreTraining.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput" | |
| >FunnelForPreTrainingOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.models.funnel.modeling_funnel.FunnelForPreTrainingOutput" | |
| >FunnelForPreTrainingOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var ee=e($,4);u(ee,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var qe=e(ee,4);m(qe,{anchor:"transformers.FunnelForPreTraining.forward.example",children:(o,r)=>{var n=Ke(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForPreTraining | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForPreTraining.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"Hello, my dog is cute"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span>logits = model(**inputs).logits`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(K),a(F);var ne=e(F,2);d(ne,{title:"FunnelForMaskedLM",local:"transformers.FunnelForMaskedLM",headingTag:"h2"});var v=e(ne,2),oe=s(v);t(oe,{name:"class transformers.FunnelForMaskedLM",anchor:"transformers.FunnelForMaskedLM",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L994",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelForMaskedLM.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var te=e(oe,8),se=s(te);t(se,{name:"forward",anchor:"transformers.FunnelForMaskedLM.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1012",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelForMaskedLM.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelForMaskedLM.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelForMaskedLM.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelForMaskedLM.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelForMaskedLM.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Labels for computing the masked language modeling loss. Indices should be in <code>[-100, 0, ..., config.vocab_size]</code> (see <code>input_ids</code> docstring) Tokens with indices set to <code>-100</code> are ignored (masked), the | |
| loss is only computed for the tokens with labels in <code>[0, ..., config.vocab_size]</code>.`,name:"labels"},{anchor:"transformers.FunnelForMaskedLM.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelForMaskedLM.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelForMaskedLM.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput" | |
| >MaskedLMOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput" | |
| >MaskedLMOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var ae=e(se,4);u(ae,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var Ne=e(ae,4);m(Ne,{anchor:"transformers.FunnelForMaskedLM.forward.example",children:(o,r)=>{var n=_(),i=e(c(n),2);p(i,{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Ub2tlbml6ZXIlMkMlMjBGdW5uZWxGb3JNYXNrZWRMTSUwQWltcG9ydCUyMHRvcmNoJTBBJTBBdG9rZW5pemVyJTIwJTNEJTIwQXV0b1Rva2VuaXplci5mcm9tX3ByZXRyYWluZWQoJTIyZnVubmVsLXRyYW5zZm9ybWVyJTJGc21hbGwlMjIpJTBBbW9kZWwlMjAlM0QlMjBGdW5uZWxGb3JNYXNrZWRMTS5mcm9tX3ByZXRyYWluZWQoJTIyZnVubmVsLXRyYW5zZm9ybWVyJTJGc21hbGwlMjIpJTBBJTBBaW5wdXRzJTIwJTNEJTIwdG9rZW5pemVyKCUyMlRoZSUyMGNhcGl0YWwlMjBvZiUyMEZyYW5jZSUyMGlzJTIwJTNDbWFzayUzRS4lMjIlMkMlMjByZXR1cm5fdGVuc29ycyUzRCUyMnB0JTIyKSUwQSUwQXdpdGglMjB0b3JjaC5ub19ncmFkKCklM0ElMEElMjAlMjAlMjAlMjBsb2dpdHMlMjAlM0QlMjBtb2RlbCgqKmlucHV0cykubG9naXRzJTBBJTBBJTIzJTIwcmV0cmlldmUlMjBpbmRleCUyMG9mJTIwJTNDbWFzayUzRSUwQW1hc2tfdG9rZW5faW5kZXglMjAlM0QlMjAoaW5wdXRzLmlucHV0X2lkcyUyMCUzRCUzRCUyMHRva2VuaXplci5tYXNrX3Rva2VuX2lkKSU1QjAlNUQubm9uemVybyhhc190dXBsZSUzRFRydWUpJTVCMCU1RCUwQSUwQXByZWRpY3RlZF90b2tlbl9pZCUyMCUzRCUyMGxvZ2l0cyU1QjAlMkMlMjBtYXNrX3Rva2VuX2luZGV4JTVELmFyZ21heChheGlzJTNELTEpJTBBdG9rZW5pemVyLmRlY29kZShwcmVkaWN0ZWRfdG9rZW5faWQpJTBBJTBBbGFiZWxzJTIwJTNEJTIwdG9rZW5pemVyKCUyMlRoZSUyMGNhcGl0YWwlMjBvZiUyMEZyYW5jZSUyMGlzJTIwUGFyaXMuJTIyJTJDJTIwcmV0dXJuX3RlbnNvcnMlM0QlMjJwdCUyMiklNUIlMjJpbnB1dF9pZHMlMjIlNUQlMEElMjMlMjBtYXNrJTIwbGFiZWxzJTIwb2YlMjBub24tJTNDbWFzayUzRSUyMHRva2VucyUwQWxhYmVscyUyMCUzRCUyMHRvcmNoLndoZXJlKGlucHV0cy5pbnB1dF9pZHMlMjAlM0QlM0QlMjB0b2tlbml6ZXIubWFza190b2tlbl9pZCUyQyUyMGxhYmVscyUyQyUyMC0xMDApJTBBJTBBb3V0cHV0cyUyMCUzRCUyMG1vZGVsKCoqaW5wdXRzJTJDJTIwbGFiZWxzJTNEbGFiZWxzKSUwQXJvdW5kKG91dHB1dHMubG9zcy5pdGVtKCklMkMlMjAyKQ==",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForMaskedLM | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForMaskedLM.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"The capital of France is <mask>."</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># retrieve index of <mask></span> | |
| <span class="hljs-meta">>>> </span>mask_token_index = (inputs.input_ids == tokenizer.mask_token_id)[<span class="hljs-number">0</span>].nonzero(as_tuple=<span class="hljs-literal">True</span>)[<span class="hljs-number">0</span>] | |
| <span class="hljs-meta">>>> </span>predicted_token_id = logits[<span class="hljs-number">0</span>, mask_token_index].argmax(axis=-<span class="hljs-number">1</span>) | |
| <span class="hljs-meta">>>> </span>tokenizer.decode(predicted_token_id) | |
| ... | |
| <span class="hljs-meta">>>> </span>labels = tokenizer(<span class="hljs-string">"The capital of France is Paris."</span>, return_tensors=<span class="hljs-string">"pt"</span>)[<span class="hljs-string">"input_ids"</span>] | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># mask labels of non-<mask> tokens</span> | |
| <span class="hljs-meta">>>> </span>labels = torch.where(inputs.input_ids == tokenizer.mask_token_id, labels, -<span class="hljs-number">100</span>) | |
| <span class="hljs-meta">>>> </span>outputs = model(**inputs, labels=labels) | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(outputs.loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(te),a(v);var re=e(v,2);d(re,{title:"FunnelForSequenceClassification",local:"transformers.FunnelForSequenceClassification",headingTag:"h2"});var j=e(re,2),le=s(j);t(le,{name:"class transformers.FunnelForSequenceClassification",anchor:"transformers.FunnelForSequenceClassification",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1063",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelForSequenceClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var ie=e(le,8),de=s(ie);t(de,{name:"forward",anchor:"transformers.FunnelForSequenceClassification.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1074",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelForSequenceClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelForSequenceClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelForSequenceClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelForSequenceClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelForSequenceClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for computing the sequence classification/regression loss. Indices should be in <code>[0, ..., config.num_labels - 1]</code>. If <code>config.num_labels == 1</code> a regression loss is computed (Mean-Square loss), If | |
| <code>config.num_labels > 1</code> a classification loss is computed (Cross-Entropy).`,name:"labels"},{anchor:"transformers.FunnelForSequenceClassification.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelForSequenceClassification.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelForSequenceClassification.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput" | |
| >SequenceClassifierOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput" | |
| >SequenceClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var ce=e(de,4);u(ce,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var pe=e(ce,4);m(pe,{anchor:"transformers.FunnelForSequenceClassification.forward.example",children:(o,r)=>{var n=$e(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForSequenceClassification | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForSequenceClassification.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"Hello, my dog is cute"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span>predicted_class_id = logits.argmax().item() | |
| <span class="hljs-meta">>>> </span>model.config.id2label[predicted_class_id] | |
| ... | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># To train a model on \`num_labels\` classes, you can pass \`num_labels=num_labels\` to \`.from_pretrained(...)\`</span> | |
| <span class="hljs-meta">>>> </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label) | |
| <span class="hljs-meta">>>> </span>model = FunnelForSequenceClassification.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>, num_labels=num_labels) | |
| <span class="hljs-meta">>>> </span>labels = torch.tensor([<span class="hljs-number">1</span>]) | |
| <span class="hljs-meta">>>> </span>loss = model(**inputs, labels=labels).loss | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}});var Ie=e(pe,2);m(Ie,{anchor:"transformers.FunnelForSequenceClassification.forward.example-2",children:(o,r)=>{var n=en(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForSequenceClassification | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForSequenceClassification.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>, problem_type=<span class="hljs-string">"multi_label_classification"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(<span class="hljs-string">"Hello, my dog is cute"</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span>predicted_class_ids = torch.arange(<span class="hljs-number">0</span>, logits.shape[-<span class="hljs-number">1</span>])[torch.sigmoid(logits).squeeze(dim=<span class="hljs-number">0</span>) > <span class="hljs-number">0.5</span>] | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># To train a model on \`num_labels\` classes, you can pass \`num_labels=num_labels\` to \`.from_pretrained(...)\`</span> | |
| <span class="hljs-meta">>>> </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label) | |
| <span class="hljs-meta">>>> </span>model = FunnelForSequenceClassification.from_pretrained( | |
| <span class="hljs-meta">... </span> <span class="hljs-string">"funnel-transformer/small"</span>, num_labels=num_labels, problem_type=<span class="hljs-string">"multi_label_classification"</span> | |
| <span class="hljs-meta">... </span>) | |
| <span class="hljs-meta">>>> </span>labels = torch.<span class="hljs-built_in">sum</span>( | |
| <span class="hljs-meta">... </span> torch.nn.functional.one_hot(predicted_class_ids[<span class="hljs-literal">None</span>, :].clone(), num_classes=num_labels), dim=<span class="hljs-number">1</span> | |
| <span class="hljs-meta">... </span>).to(torch.<span class="hljs-built_in">float</span>) | |
| <span class="hljs-meta">>>> </span>loss = model(**inputs, labels=labels).loss`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(ie),a(j);var ue=e(j,2);d(ue,{title:"FunnelForMultipleChoice",local:"transformers.FunnelForMultipleChoice",headingTag:"h2"});var z=e(ue,2),me=s(z);t(me,{name:"class transformers.FunnelForMultipleChoice",anchor:"transformers.FunnelForMultipleChoice",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1139",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelForMultipleChoice.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var he=e(me,8),fe=s(he);t(fe,{name:"forward",anchor:"transformers.FunnelForMultipleChoice.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1148",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelForMultipleChoice.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelForMultipleChoice.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelForMultipleChoice.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelForMultipleChoice.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelForMultipleChoice.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for computing the multiple choice classification loss. Indices should be in <code>[0, ..., num_choices-1]</code> where <code>num_choices</code> is the size of the second dimension of the input tensors. (See | |
| <code>input_ids</code> above)`,name:"labels"},{anchor:"transformers.FunnelForMultipleChoice.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelForMultipleChoice.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelForMultipleChoice.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.MultipleChoiceModelOutput" | |
| >MultipleChoiceModelOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.MultipleChoiceModelOutput" | |
| >MultipleChoiceModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var ge=e(fe,4);u(ge,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var Ge=e(ge,4);m(Ge,{anchor:"transformers.FunnelForMultipleChoice.forward.example",children:(o,r)=>{var n=_(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForMultipleChoice | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForMultipleChoice.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>prompt = <span class="hljs-string">"In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced."</span> | |
| <span class="hljs-meta">>>> </span>choice0 = <span class="hljs-string">"It is eaten with a fork and a knife."</span> | |
| <span class="hljs-meta">>>> </span>choice1 = <span class="hljs-string">"It is eaten while held in the hand."</span> | |
| <span class="hljs-meta">>>> </span>labels = torch.tensor(<span class="hljs-number">0</span>).unsqueeze(<span class="hljs-number">0</span>) <span class="hljs-comment"># choice0 is correct (according to Wikipedia ;)), batch size 1</span> | |
| <span class="hljs-meta">>>> </span>encoding = tokenizer([prompt, prompt], [choice0, choice1], return_tensors=<span class="hljs-string">"pt"</span>, padding=<span class="hljs-literal">True</span>) | |
| <span class="hljs-meta">>>> </span>outputs = model(**{k: v.unsqueeze(<span class="hljs-number">0</span>) <span class="hljs-keyword">for</span> k, v <span class="hljs-keyword">in</span> encoding.items()}, labels=labels) <span class="hljs-comment"># batch size is 1</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># the linear classifier still needs to be trained</span> | |
| <span class="hljs-meta">>>> </span>loss = outputs.loss | |
| <span class="hljs-meta">>>> </span>logits = outputs.logits`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(he),a(z);var _e=e(z,2);d(_e,{title:"FunnelForTokenClassification",local:"transformers.FunnelForTokenClassification",headingTag:"h2"});var J=e(_e,2),be=s(J);t(be,{name:"class transformers.FunnelForTokenClassification",anchor:"transformers.FunnelForTokenClassification",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1206",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelForTokenClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var ye=e(be,8),Te=s(ye);t(Te,{name:"forward",anchor:"transformers.FunnelForTokenClassification.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1218",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelForTokenClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelForTokenClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelForTokenClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelForTokenClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelForTokenClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Labels for computing the token classification loss. Indices should be in <code>[0, ..., config.num_labels - 1]</code>.`,name:"labels"},{anchor:"transformers.FunnelForTokenClassification.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelForTokenClassification.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelForTokenClassification.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput" | |
| >TokenClassifierOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput" | |
| >TokenClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var Me=e(Te,4);u(Me,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var Re=e(Me,4);m(Re,{anchor:"transformers.FunnelForTokenClassification.forward.example",children:(o,r)=>{var n=_(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForTokenClassification | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForTokenClassification.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer( | |
| <span class="hljs-meta">... </span> <span class="hljs-string">"HuggingFace is a company based in Paris and New York"</span>, add_special_tokens=<span class="hljs-literal">False</span>, return_tensors=<span class="hljs-string">"pt"</span> | |
| <span class="hljs-meta">... </span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> logits = model(**inputs).logits | |
| <span class="hljs-meta">>>> </span>predicted_token_class_ids = logits.argmax(-<span class="hljs-number">1</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Note that tokens are classified rather then input words which means that</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># there might be more predicted token classes than words.</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># Multiple token classes might account for the same word</span> | |
| <span class="hljs-meta">>>> </span>predicted_tokens_classes = [model.config.id2label[t.item()] <span class="hljs-keyword">for</span> t <span class="hljs-keyword">in</span> predicted_token_class_ids[<span class="hljs-number">0</span>]] | |
| <span class="hljs-meta">>>> </span>predicted_tokens_classes | |
| ... | |
| <span class="hljs-meta">>>> </span>labels = predicted_token_class_ids | |
| <span class="hljs-meta">>>> </span>loss = model(**inputs, labels=labels).loss | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(ye),a(J);var ke=e(J,2);d(ke,{title:"FunnelForQuestionAnswering",local:"transformers.FunnelForQuestionAnswering",headingTag:"h2"});var U=e(ke,2),we=s(U);t(we,{name:"class transformers.FunnelForQuestionAnswering",anchor:"transformers.FunnelForQuestionAnswering",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1265",parameters:[{name:"config",val:": FunnelConfig"}],parametersDescription:[{anchor:"transformers.FunnelForQuestionAnswering.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig">FunnelConfig</a>) — | |
| Model configuration class with all the parameters of the model. Initializing with a config file does not | |
| load the weights associated with the model, only the configuration. Check out the | |
| <a href="/docs/transformers/pr_36708/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var Fe=e(we,8),ve=s(Fe);t(ve,{name:"forward",anchor:"transformers.FunnelForQuestionAnswering.forward",source:"https://github.com/huggingface/transformers/blob/vr_36708/src/transformers/models/funnel/modeling_funnel.py#L1276",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"start_positions",val:": typing.Optional[torch.Tensor] = None"},{name:"end_positions",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FunnelForQuestionAnswering.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p> | |
| <p>Indices can be obtained using <a href="/docs/transformers/pr_36708/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and | |
| <a href="/docs/transformers/pr_36708/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p> | |
| <p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.FunnelForQuestionAnswering.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>1 for tokens that are <strong>not masked</strong>,</li> | |
| <li>0 for tokens that are <strong>masked</strong>.</li> | |
| </ul> | |
| <p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.FunnelForQuestionAnswering.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) — | |
| Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p> | |
| <ul> | |
| <li>0 corresponds to a <em>sentence A</em> token,</li> | |
| <li>1 corresponds to a <em>sentence B</em> token.</li> | |
| </ul> | |
| <p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.FunnelForQuestionAnswering.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) — | |
| Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This | |
| is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the | |
| model’s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.FunnelForQuestionAnswering.forward.start_positions",description:`<strong>start_positions</strong> (<code>torch.Tensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for position (index) of the start of the labelled span for computing the token classification loss. | |
| Positions are clamped to the length of the sequence (<code>sequence_length</code>). Position outside of the sequence | |
| are not taken into account for computing the loss.`,name:"start_positions"},{anchor:"transformers.FunnelForQuestionAnswering.forward.end_positions",description:`<strong>end_positions</strong> (<code>torch.Tensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) — | |
| Labels for position (index) of the end of the labelled span for computing the token classification loss. | |
| Positions are clamped to the length of the sequence (<code>sequence_length</code>). Position outside of the sequence | |
| are not taken into account for computing the loss.`,name:"end_positions"},{anchor:"transformers.FunnelForQuestionAnswering.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned | |
| tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.FunnelForQuestionAnswering.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for | |
| more detail.`,name:"output_hidden_states"},{anchor:"transformers.FunnelForQuestionAnswering.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) — | |
| Whether or not to return a <a href="/docs/transformers/pr_36708/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p>A <a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput" | |
| >QuestionAnsweringModelOutput</a> or a tuple of | |
| <code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various | |
| elements depending on the configuration (<a | |
| href="/docs/transformers/pr_36708/en/model_doc/funnel#transformers.FunnelConfig" | |
| >FunnelConfig</a>) and inputs.</p> | |
| `,returnType:`<script context="module">export const metadata = 'undefined';<\/script> | |
| <p><a | |
| href="/docs/transformers/pr_36708/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput" | |
| >QuestionAnsweringModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p> | |
| `});var je=e(ve,4);u(je,{children:(o,r)=>{var n=h();l(o,n)},$$slots:{default:!0}});var Xe=e(je,4);m(Xe,{anchor:"transformers.FunnelForQuestionAnswering.forward.example",children:(o,r)=>{var n=_(),i=e(c(n),2);p(i,{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FunnelForQuestionAnswering | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>model = FunnelForQuestionAnswering.from_pretrained(<span class="hljs-string">"funnel-transformer/small"</span>) | |
| <span class="hljs-meta">>>> </span>question, text = <span class="hljs-string">"Who was Jim Henson?"</span>, <span class="hljs-string">"Jim Henson was a nice puppet"</span> | |
| <span class="hljs-meta">>>> </span>inputs = tokenizer(question, text, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> outputs = model(**inputs) | |
| <span class="hljs-meta">>>> </span>answer_start_index = outputs.start_logits.argmax() | |
| <span class="hljs-meta">>>> </span>answer_end_index = outputs.end_logits.argmax() | |
| <span class="hljs-meta">>>> </span>predict_answer_tokens = inputs.input_ids[<span class="hljs-number">0</span>, answer_start_index : answer_end_index + <span class="hljs-number">1</span>] | |
| <span class="hljs-meta">>>> </span>tokenizer.decode(predict_answer_tokens, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| ... | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># target is "nice puppet"</span> | |
| <span class="hljs-meta">>>> </span>target_start_index = torch.tensor([<span class="hljs-number">14</span>]) | |
| <span class="hljs-meta">>>> </span>target_end_index = torch.tensor([<span class="hljs-number">15</span>]) | |
| <span class="hljs-meta">>>> </span>outputs = model(**inputs, start_positions=target_start_index, end_positions=target_end_index) | |
| <span class="hljs-meta">>>> </span>loss = outputs.loss | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>) | |
| ...`,lang:"python",wrap:!1}),l(o,n)},$$slots:{default:!0}}),a(Fe),a(U);var Ve=e(U,2);Ye(Ve,{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/funnel.md"}),g(2),l(ze,C),Le()}export{ln as component}; | |
Xet Storage Details
- Size:
- 130 kB
- Xet hash:
- 48c79a9d29c3b0f00990df1951fc55db73fd55cbc65fa866294fc63db4307ace
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.