SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[Doctests] Move doctests to new GPU & Fix bugs (#15969)

Browse Source 
* test

* up

* up

* Empty test commit

* up

* update tests

* up

* fix some vision models

* correct

* correct docs

* Trigger notification

* finalize

* check

* correct quicktour

* Apply suggestions from code review

* improve doctests

* Trigger Build

* next try

* next try

* and again

* Output current clone information

* Output current clone information

* Correct path

* add tf round again

* revert to daily job

Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
This commit is contained in:
Patrick von Platen
2022-03-09 13:09:56 +01:00
committed by GitHub
parent f4e4ad34cc
commit c1aaa43935
12 changed files with 50 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
.github/workflows/doctests.yml vendored
View File

@@ -16,35 +16,43 @@ env:
OMP_NUM_THREADS: 16
MKL_NUM_THREADS: 16
PYTEST_TIMEOUT: 600
SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
TF_FORCE_GPU_ALLOW_GROWTH: true
jobs:
run_doctests:
runs-on: [self-hosted, docker-gpu-test, single-gpu]
runs-on: [self-hosted, doc-tests-gpu]
container:
image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
image: huggingface/transformers-all-latest-gpu
options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
steps:
- name: Launcher docker
uses: actions/checkout@v2
- uses: actions/checkout@v2
with:
repository: 'huggingface/transformers'
path: transformers
- name: NVIDIA-SMI
run: |
nvidia-smi
- name: Install dependencies
- name: GPU visibility
working-directory: transformers
run: |
apt -y update && apt install -y libsndfile1-dev
pip install --upgrade pip
pip install .[testing,torch-speech]
utils/print_env_pt.py
TF_CPP_MIN_LOG_LEVEL=3 python3 -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
TF_CPP_MIN_LOG_LEVEL=3 python3 -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
- name: Prepare files for doctests
working-directory: transformers
run: |
python utils/prepare_for_doc_test.py src docs
python3 utils/prepare_for_doc_test.py src docs
- name: Run doctests
working-directory: transformers
run: |
pytest --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.mdx"
python3 -m pytest --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.mdx"
- name: Clean files after doctests
working-directory: transformers
run: |
python utils/prepare_for_doc_test.py src docs --remove_new_line
python3 utils/prepare_for_doc_test.py src docs --remove_new_line

11
docs/source/quicktour.mdx
View File

@@ -99,12 +99,13 @@ The [`pipeline`] can also iterate over an entire dataset. Start by installing th
pip install datasets
```
Create a [`pipeline`] with the task you want to solve for and the model you want to use. Set the `device` parameter to `0` to place the tensors on a CUDA device:
Create a [`pipeline`] with the task you want to solve for and the model you want to use.
```py
>>> import torch
>>> from transformers import pipeline
>>> speech_recognizer = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h", device=0)
>>> speech_recognizer = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")
```
Next, load a dataset (see the 🤗 Datasets [Quick Start](https://huggingface.co/docs/datasets/quickstart.html) for more details) you'd like to iterate over. For example, let's load the [SUPERB](https://huggingface.co/datasets/superb) dataset:
@@ -264,10 +265,10 @@ tensor([[0.0021, 0.0018, 0.0115, 0.2121, 0.7725],
>>> import tensorflow as tf
>>> tf_predictions = tf.nn.softmax(tf_outputs.logits, axis=-1)
>>> print(tf_predictions)
>>> print(tf.math.round(tf_predictions * 10**4) / 10**4)
tf.Tensor(
[[0.00206 0.00177 0.01155 0.21209 0.77253]
[0.20842 0.18262 0.19693 0.1755 0.23652]], shape=(2, 5), dtype=float32)
[[0.0021 0.0018 0.0116 0.2121 0.7725]
[0.2084 0.1826 0.1969 0.1755 0.2365]], shape=(2, 5), dtype=float32)
```
<Tip>

2
src/transformers/models/beit/modeling_beit.py
View File

@@ -55,7 +55,7 @@ _EXPECTED_OUTPUT_SHAPE = [1, 197, 768]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "microsoft/beit-base-patch16-224"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'tabby, tabby cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
BEIT_PRETRAINED_MODEL_ARCHIVE_LIST = [
"microsoft/beit-base-patch16-224",

2
src/transformers/models/convnext/modeling_convnext.py
View File

@@ -46,7 +46,7 @@ _EXPECTED_OUTPUT_SHAPE = [1, 768, 7, 7]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "facebook/convnext-tiny-224"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'tabby, tabby cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
CONVNEXT_PRETRAINED_MODEL_ARCHIVE_LIST = [
"facebook/convnext-tiny-224",

5
src/transformers/models/deit/modeling_deit.py
View File

@@ -51,7 +51,7 @@ _EXPECTED_OUTPUT_SHAPE = [1, 198, 768]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "facebook/deit-base-distilled-patch16-224"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'tabby, tabby cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
DEIT_PRETRAINED_MODEL_ARCHIVE_LIST = [
@@ -697,9 +697,11 @@ class DeiTForImageClassification(DeiTPreTrainedModel):
```python
>>> from transformers import DeiTFeatureExtractor, DeiTForImageClassification
>>> import torch
>>> from PIL import Image
>>> import requests
>>> torch.manual_seed(3) # doctest: +IGNORE_RESULT
>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
>>> image = Image.open(requests.get(url, stream=True).raw)
@@ -714,6 +716,7 @@ class DeiTForImageClassification(DeiTPreTrainedModel):
>>> # model predicts one of the 1000 ImageNet classes
>>> predicted_class_idx = logits.argmax(-1).item()
>>> print("Predicted class:", model.config.id2label[predicted_class_idx])
Predicted class: maillot
```"""
return_dict = return_dict if return_dict is not None else self.config.use_return_dict

4
src/transformers/models/poolformer/modeling_poolformer.py
View File

@@ -44,11 +44,11 @@ _FEAT_EXTRACTOR_FOR_DOC = "PoolFormerFeatureExtractor"
# Base docstring
_CHECKPOINT_FOR_DOC = "sail/poolformer_s12"
_EXPECTED_OUTPUT_SHAPE = [1, 197, 768]
_EXPECTED_OUTPUT_SHAPE = [1, 512, 7, 7]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "sail/poolformer_s12"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'tabby, tabby cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
POOLFORMER_PRETRAINED_MODEL_ARCHIVE_LIST = [
"sail/poolformer_s12",

2
src/transformers/models/segformer/modeling_segformer.py
View File

@@ -49,7 +49,7 @@ _EXPECTED_OUTPUT_SHAPE = [1, 256, 16, 16]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "nvidia/mit-b0"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'tabby, tabby cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
SEGFORMER_PRETRAINED_MODEL_ARCHIVE_LIST = [
"nvidia/segformer-b0-finetuned-ade-512-512",

10
src/transformers/models/speech_to_text/modeling_speech_to_text.py
View File

@@ -1168,9 +1168,10 @@ class Speech2TextModel(Speech2TextPreTrainedModel):
>>> model = Speech2TextModel.from_pretrained("facebook/s2t-small-librispeech-asr")
>>> feature_extractor = Speech2TextFeatureExtractor.from_pretrained("facebook/s2t-small-librispeech-asr")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> input_features = feature_extractor(
>>> inputs = feature_extractor(
... ds[0]["audio"]["array"], sampling_rate=ds[0]["audio"]["sampling_rate"], return_tensors="pt"
>>> ).input_features
... )
>>> input_features = inputs.input_features
>>> decoder_input_ids = torch.tensor([[1, 1]]) * model.config.decoder_start_token_id
>>> last_hidden_state = model(input_features, decoder_input_ids=decoder_input_ids).last_hidden_state
>>> list(last_hidden_state.shape)
@@ -1322,9 +1323,10 @@ class Speech2TextForConditionalGeneration(Speech2TextPreTrainedModel):
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> input_features = processor(
>>> inputs = processor(
... ds[0]["audio"]["array"], sampling_rate=ds[0]["audio"]["sampling_rate"], return_tensors="pt"
>>> ).input_features
... )
>>> input_features = inputs.input_features
>>> generated_ids = model.generate(inputs=input_features)

15
src/transformers/models/speech_to_text_2/modeling_speech_to_text_2.py
View File

@@ -874,24 +874,25 @@ class Speech2Text2ForCausalLM(Speech2Text2PreTrainedModel):
>>> encoder = Wav2Vec2Model(Wav2Vec2Config())
>>> decoder = Speech2Text2ForCausalLM(Speech2Text2Config())
# init random speech2text model
>>> # init random speech2text model
>>> model = SpeechEncoderDecoderModel(encoder=encoder, decoder=decoder)
>>> model.config.pad_token_id = tokenizer.pad_token_id
>>> model.config.decoder_start_token_id = tokenizer.bos_token_id
# pre-process inputs and labels
>>> # pre-process inputs and labels
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> input_values = feature_extractor(
>>> inputs = feature_extractor(
... ds[0]["audio"]["array"], sampling_rate=ds[0]["audio"]["sampling_rate"], return_tensors="pt"
>>> ).input_values # Batch size 1
... )
>>> input_values = inputs.input_values
>>> decoder_input_ids = tokenizer(ds[0]["text"], return_tensors="pt").input_ids
# compute loss
>>> # compute loss
>>> loss = model(inputs=input_values, labels=decoder_input_ids).loss
# backprop loss
>>> # backprop loss
>>> loss.backward()
>>> loss.backward() # doctest: +IGNORE_RESULT
```"""
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions

2
src/transformers/models/swin/modeling_swin.py
View File

@@ -48,7 +48,7 @@ _EXPECTED_OUTPUT_SHAPE = [1, 49, 768]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "microsoft/swin-tiny-patch4-window7-224"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'tabby, tabby cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "tabby, tabby cat"
SWIN_PRETRAINED_MODEL_ARCHIVE_LIST = [

2
src/transformers/models/vit/modeling_vit.py
View File

@@ -48,7 +48,7 @@ _EXPECTED_OUTPUT_SHAPE = [1, 197, 768]
# Image classification docstring
_IMAGE_CLASS_CHECKPOINT = "google/vit-base-patch16-224"
_IMAGE_CLASS_EXPECTED_OUTPUT = "'Egyptian cat'"
_IMAGE_CLASS_EXPECTED_OUTPUT = "Egyptian cat"
VIT_PRETRAINED_MODEL_ARCHIVE_LIST = [

35
src/transformers/models/wav2vec2/modeling_wav2vec2.py
View File

@@ -1611,7 +1611,6 @@ class Wav2Vec2ForMaskedLM(Wav2Vec2PreTrainedModel):
self.post_init()
@add_start_docstrings_to_model_forward(WAV_2_VEC_2_INPUTS_DOCSTRING)
@replace_return_docstrings(output_type=Wav2Vec2BaseModelOutput, config_class=_CONFIG_FOR_DOC)
def forward(
self,
input_values,
@@ -1621,40 +1620,6 @@ class Wav2Vec2ForMaskedLM(Wav2Vec2PreTrainedModel):
return_dict=None,
labels=None,
):
r"""
labels (`torch.LongTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
TODO(PVP): Fill out when adding training
Returns:
Example:
```python
>>> from transformers import Wav2Vec2Processor, Wav2Vec2ForMaskedLM
>>> from datasets import load_dataset
>>> import soundfile as sf
>>> import torch
>>> processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
>>> model = Wav2Vec2ForMaskedLM.from_pretrained("facebook/wav2vec2-base-960h")
>>> def map_to_array(batch):
... speech, _ = sf.read(batch["file"])
... batch["speech"] = speech
... return batch
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> ds = ds.map(map_to_array)
>>> input_values = processor(ds["speech"][0], return_tensors="pt").input_values # Batch size 1
>>> logits = model(input_values).logits
>>> predicted_ids = torch.argmax(logits, dim=-1)
>>> transcription = processor.decode(predicted_ids[0])
```"""
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.wav2vec2(