Merge branch 'master' into add_models_special_tokens_to_specific_configs

.circleci/config.yml

@@ -14,22 +14,6 @@ jobs:
             - run: sudo pip install codecov pytest-cov
             - run: python -m pytest -n 8 --dist=loadfile -s -v ./tests/ --cov
             - run: codecov
-    run_all_tests_torch_and_tf:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.5
-        environment:
-            OMP_NUM_THREADS: 1
-            RUN_SLOW: yes
-            RUN_CUSTOM_TOKENIZERS: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - run: sudo pip install .[mecab,sklearn,tf-cpu,torch,testing]
-            - run:
-                command: python -m pytest -n 8 --dist=loadfile -s -v ./tests/
-                no_output_timeout: 4h
 
     run_tests_torch:
         working_directory: ~/transformers
@@ -134,13 +118,3 @@ workflows:
             - run_tests_torch
             - run_tests_tf
             - deploy_doc: *workflow_filters
-    run_slow_tests:
-        triggers:
-            - schedule:
-                cron: "0 4 * * *"
-                filters:
-                    branches:
-                        only:
-                            - master
-        jobs:
-            - run_all_tests_torch_and_tf

.github/workflows/github-push.yml

+name: GitHub-hosted runner
+
+on: push
+
+jobs:
+  check_code_quality:
+    runs-on: ubuntu-18.04
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python
+      uses: actions/setup-python@v1
+      with:
+        python-version: 3.7
+    - name: Install dependencies
+      run: |
+        pip install .[tf,torch,quality]
+
+
+

.github/workflows/self-push.yml

+name: Self-hosted runner (push)
+
+on: 
+  push:
+    branches:
+      - master
+  pull_request:
+
+
+jobs:
+  run_tests_torch_and_tf_gpu:
+    runs-on: self-hosted
+    steps:
+    - uses: actions/checkout@v2
+    - name: Python version
+      run: |
+        which python
+        python --version
+        pip --version
+    - name: Current dir
+      run: pwd
+    - run: nvidia-smi
+    - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
+      run: |
+        python -m venv .env
+        source .env/bin/activate
+        which python
+        python --version
+        pip --version
+    - name: Install dependencies
+      run: |
+        source .env/bin/activate
+        pip install .[sklearn,tf,torch,testing]
+
+    - name: Are GPUs recognized by our DL frameworks
+      run: |
+        source .env/bin/activate
+        python -c "import torch; print(torch.cuda.is_available())"
+        python -c "import tensorflow as tf; print(tf.test.is_built_with_cuda(), tf.config.list_physical_devices('GPU'))"
+
+    - name: Run all non-slow tests on GPU
+      env:
+        TF_FORCE_GPU_ALLOW_GROWTH: "true"
+        # TF_GPU_MEMORY_LIMIT: 4096
+        OMP_NUM_THREADS: 1
+        USE_CUDA: yes
+      run: |
+        source .env/bin/activate
+        python -m pytest -n 2 --dist=loadfile -s -v ./tests/

.github/workflows/self-scheduled.yml

+name: Self-hosted runner (scheduled)
+
+on:
+  push:
+    branches:
+      - ci_*
+  repository_dispatch:
+  schedule:
+    - cron: "0 0 * * *"
+
+jobs:
+  run_all_tests_torch_and_tf_gpu:
+    runs-on: self-hosted
+    steps:
+    - uses: actions/checkout@v2
+    - name: Python version
+      run: |
+        which python
+        python --version
+        pip --version
+    - name: Current dir
+      run: pwd
+    - run: nvidia-smi
+    - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
+      run: |
+        python -m venv .env
+        source .env/bin/activate
+        which python
+        python --version
+        pip --version
+    - name: Install dependencies
+      run: |
+        source .env/bin/activate
+        pip install .[sklearn,tf,torch,testing]
+
+    - name: Are GPUs recognized by our DL frameworks
+      run: |
+        source .env/bin/activate
+        python -c "import torch; print(torch.cuda.is_available())"
+        python -c "import tensorflow as tf; print(tf.test.is_built_with_cuda(), tf.config.list_physical_devices('GPU'))"
+
+    - name: Run all tests on GPU
+      env:
+        TF_FORCE_GPU_ALLOW_GROWTH: "true"
+        OMP_NUM_THREADS: 1
+        RUN_SLOW: yes
+        USE_CUDA: yes
+      run: |
+        source .env/bin/activate
+        python -m pytest -n 1 --dist=loadfile -s -v ./tests/
+        

docker/Dockerfile

-FROM pytorch/pytorch:latest
-
-RUN git clone https://github.com/NVIDIA/apex.git && cd apex && python setup.py install --cuda_ext --cpp_ext
-
-RUN pip install transformers
-
-WORKDIR /workspace
\ No newline at end of file

docker/transformers-cpu/Dockerfile

+FROM ubuntu:18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    jupyter \
+    tensorflow-cpu \
+    torch
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]
\ No newline at end of file

docker/transformers-gpu/Dockerfile

+FROM nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    jupyter \
+    tensorflow \
+    torch
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]
\ No newline at end of file

docker/transformers-pytorch-cpu/Dockerfile

+FROM ubuntu:18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    jupyter \
+    torch
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]
\ No newline at end of file

docker/transformers-pytorch-gpu/Dockerfile

+FROM nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    mkl \
+    torch
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]
\ No newline at end of file

docker/transformers-tensorflow-cpu/Dockerfile

+FROM ubuntu:18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    mkl \
+    tensorflow-cpu
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]
\ No newline at end of file

docker/transformers-tensorflow-gpu/Dockerfile

+FROM nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    mkl \
+    tensorflow
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]
\ No newline at end of file

docs/source/_static/css/huggingface.css

+/* Our DOM objects */
+
+.framework-selector {
+    display: flex;
+    flex-direction: row;
+    justify-content: flex-end;
+}
+
+.framework-selector > button {
+    background-color: white;
+    color: #6670FF;
+    border: 1px solid #6670FF;
+    padding: 5px;
+}
+
+.framework-selector > button.selected{
+    background-color: #6670FF;
+    color: white;
+    border: 1px solid #6670FF;
+    padding: 5px;
+}
+
 /* The literal code blocks */
 .rst-content tt.literal, .rst-content tt.literal, .rst-content code.literal {
     color: #6670FF;

docs/source/_static/js/custom.js

@@ -68,6 +68,74 @@ function addHfMenu() {
     document.body.insertAdjacentHTML('afterbegin', div);
 }
 
+function platformToggle() {
+    const codeBlocks = Array.from(document.getElementsByClassName("highlight"));
+    const pytorchIdentifier = "## PYTORCH CODE";
+    const tensorflowIdentifier = "## TENSORFLOW CODE";
+    const pytorchSpanIdentifier = `<span class="c1">${pytorchIdentifier}</span>`;
+    const tensorflowSpanIdentifier = `<span class="c1">${tensorflowIdentifier}</span>`;
+
+    const getFrameworkSpans = filteredCodeBlock => {
+        const spans = filteredCodeBlock.element.innerHTML;
+        const pytorchSpanPosition = spans.indexOf(pytorchSpanIdentifier);
+        const tensorflowSpanPosition = spans.indexOf(tensorflowSpanIdentifier);
+
+        let pytorchSpans;
+        let tensorflowSpans;
+
+        if(pytorchSpanPosition < tensorflowSpanPosition){
+            pytorchSpans = spans.slice(pytorchSpanPosition + pytorchSpanIdentifier.length + 1, tensorflowSpanPosition);
+            tensorflowSpans = spans.slice(tensorflowSpanPosition + tensorflowSpanIdentifier.length + 1, spans.length);
+        }else{
+            tensorflowSpans = spans.slice(tensorflowSpanPosition + tensorflowSpanIdentifier.length + 1, pytorchSpanPosition);
+            pytorchSpans = spans.slice(pytorchSpanPosition + pytorchSpanIdentifier.length + 1, spans.length);
+        }
+
+        return {
+            ...filteredCodeBlock,
+            pytorchSample: pytorchSpans ,
+            tensorflowSample: tensorflowSpans
+        }
+    };
+
+    const createFrameworkButtons = sample => {
+            const pytorchButton = document.createElement("button");
+            pytorchButton.innerText = "PyTorch";
+
+            const tensorflowButton = document.createElement("button");
+            tensorflowButton.innerText = "TensorFlow";
+
+            const selectorDiv = document.createElement("div");
+            selectorDiv.classList.add("framework-selector");
+            selectorDiv.appendChild(pytorchButton);
+            selectorDiv.appendChild(tensorflowButton);
+            sample.element.parentElement.prepend(selectorDiv);
+
+            // Init on PyTorch
+            sample.element.innerHTML = sample.pytorchSample;
+            pytorchButton.classList.add("selected");
+            tensorflowButton.classList.remove("selected");
+
+            pytorchButton.addEventListener("click", () => {
+                sample.element.innerHTML = sample.pytorchSample;
+                pytorchButton.classList.add("selected");
+                tensorflowButton.classList.remove("selected");
+            });
+            tensorflowButton.addEventListener("click", () => {
+               sample.element.innerHTML = sample.tensorflowSample;
+                tensorflowButton.classList.add("selected");
+                pytorchButton.classList.remove("selected");
+            });
+        };
+
+    codeBlocks
+        .map(element => {return {element: element.firstChild, innerText: element.innerText}})
+        .filter(codeBlock => codeBlock.innerText.includes(pytorchIdentifier) && codeBlock.innerText.includes(tensorflowIdentifier))
+        .map(getFrameworkSpans)
+        .forEach(createFrameworkButtons);
+}
+
+
 /*!
  * github-buttons v2.2.10
  * (c) 2019 なつき
@@ -85,6 +153,7 @@ function onLoad() {
     addGithubButton();
     parseGithubButtons();
     addHfMenu();
+    platformToggle();
 }
 
 window.addEventListener("load", onLoad);

docs/source/conf.py

@@ -20,7 +20,7 @@ sys.path.insert(0, os.path.abspath('../../src'))
 # -- Project information -----------------------------------------------------
 
 project = u'transformers'
-copyright = u'2019, huggingface'
+copyright = u'2020, huggingface'
 author = u'huggingface'
 
 # The short X.Y version
@@ -105,6 +105,12 @@ html_static_path = ['_static']
 #
 # html_sidebars = {}
 
+# This must be the name of an image file (path relative to the configuration 
+# directory) that is the favicon of the docs. Modern browsers use this as 
+# the icon for tabs, windows and bookmarks. It should be a Windows-style 
+# icon file (.ico).
+html_favicon = 'favicon.ico'
+
 
 # -- Options for HTMLHelp output ---------------------------------------------
 

docs/source/index.rst

@@ -61,6 +61,7 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
     quickstart
     glossary
     pretrained_models
+    usage
     model_sharing
     examples
     notebooks
@@ -79,6 +80,7 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
     main_classes/configuration
     main_classes/model
     main_classes/tokenizer
+    main_classes/pipelines
     main_classes/optimizer_schedules
     main_classes/processors
 

docs/source/main_classes/pipelines.rst

+Pipelines
+----------------------------------------------------
+
+The pipelines are a great and easy way to use models for inference. These pipelines are objects that abstract most
+of the complex code from the library, offering a simple API dedicated to several tasks, including Named Entity
+Recognition, Masked Language Modeling, Sentiment Analysis, Feature Extraction and Question Answering.
+
+There are two categories of pipeline abstractions to be aware about:
+
+- The :class:`~transformers.pipeline` which is the most powerful object encapsulating all other pipelines
+- The other task-specific pipelines, such as :class:`~transformers.NerPipeline`
+  or :class:`~transformers.QuestionAnsweringPipeline`
+
+The pipeline abstraction
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The `pipeline` abstraction is a wrapper around all the other available pipelines. It is instantiated as any
+other pipeline but requires an additional argument which is the `task`.
+
+.. autoclass:: transformers.pipeline
+    :members:
+
+
+The task specific pipelines
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Parent class: Pipeline
+=========================================
+
+.. autoclass:: transformers.Pipeline
+    :members: predict, transform, save_pretrained
+
+NerPipeline
+==========================================
+
+.. autoclass:: transformers.NerPipeline
+
+TokenClassificationPipeline
+==========================================
+
+This class is an alias of the :class:`~transformers.NerPipeline` defined above. Please refer to that pipeline for
+documentation and usage examples.
+
+FillMaskPipeline
+==========================================
+
+.. autoclass:: transformers.FillMaskPipeline
+
+FeatureExtractionPipeline
+==========================================
+
+.. autoclass:: transformers.FeatureExtractionPipeline
+
+TextClassificationPipeline
+==========================================
+
+.. autoclass:: transformers.TextClassificationPipeline
+
+QuestionAnsweringPipeline
+==========================================
+
+.. autoclass:: transformers.QuestionAnsweringPipeline
+

docs/source/model_doc/albert.rst

@@ -41,7 +41,8 @@ AlbertTokenizer
 ~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.AlbertTokenizer
-    :members:
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
 
 
 AlbertModel

docs/source/model_doc/bart.rst

@@ -4,20 +4,27 @@ Bart
 file a `Github Issue <https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title>`__ and assign
 @sshleifer
 
-The Bart model was `proposed <https://arxiv.org/abs/1910.13461>`_ by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov, Luke Zettlemoyer on 29 Oct, 2019.
-It is a sequence to sequence model where both encoder and decoder are transformers. The paper also introduces a novel pretraining objective, and demonstrates excellent summarization results.
-The authors released their code `here <https://github.com/pytorch/fairseq/tree/master/examples/bart>`_
+Paper
+~~~~~
+The Bart model was `proposed <https://arxiv.org/abs/1910.13461>`_ by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer on 29 Oct, 2019.
+According to the abstract:
 
-**Abstract:**
+- Bart uses a standard seq2seq/machine translation architecture with a bidirectional encoder (like BERT) and a left-to-right decoder (like GPT).
+- The pretraining task involves randomly shuffling the order of the original sentences and a novel in-filling scheme, where spans of text are replaced with a single mask token.
+- BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains of up to 6 ROUGE.
 
-*We present BART, a denoising autoencoder for pretraining sequence-to-sequence models. BART is trained by (1) corrupting text with an arbitrary noising function, and (2) learning a model to reconstruct the original text. It uses a standard Tranformer-based neural machine translation architecture which, despite its simplicity, can be seen as generalizing BERT (due to the bidirectional encoder), GPT (with the left-to-right decoder), and many other more recent pretraining schemes. We evaluate a number of noising approaches, finding the best performance by both randomly shuffling the order of the original sentences and using a novel in-filling scheme, where spans of text are replaced with a single mask token. BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains of up to 6 ROUGE. BART also provides a 1.1 BLEU increase over a back-translation system for machine translation, with only target language pretraining. We also report ablation experiments that replicate other pretraining schemes within the BART framework, to better measure which factors most influence end-task performance.*
-`BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension`
+The Authors' code can be found `here <https://github.com/pytorch/fairseq/tree/master/examples/bart>`_
 
 
-Notes:
+Implementation Notes
+~~~~~~~~~~~~~~~~~~~~
 - Bart doesn't use :obj:`token_type_ids`, for sequence classification just use BartTokenizer.encode to get the proper splitting.
 - Inputs to the decoder are created by BartModel.forward if they are not passed. This is different than some other model APIs.
 - Model predictions are intended to be identical to the original implementation. This only works, however, if the string you pass to fairseq.encode starts with a space.
+- Decoder inputs are created automatically by the helper function ``transformers.modeling_bart._prepare_bart_decoder_inputs``
+BartModel
+- ``MaskedLM.generate`` should be used for summarization, see the example in that docstrings
+
 
 BartModel
 ~~~~~~~~~~~~~~~~~~~~
@@ -30,7 +37,7 @@ BartForMaskedLM
 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.BartForMaskedLM
-    :members: forward
+    :members: forward, generate
 
 
 BartForSequenceClassification