Mirrors/modelscope
1
0
Fork 0
mirror of https://github.com/modelscope/modelscope.git synced 2026-07-10 04:22:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

support batch infer

Browse Source 
Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/11170755
This commit is contained in:
yichang.zyc
2022-12-28 12:17:36 +08:00
committed by yingda.chen
parent addda1f613
commit 0c79b57fcc
18 changed files with 224 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
modelscope/metrics/accuracy_metric.py
View File

@@ -40,6 +40,9 @@ class AccuracyMetric(Metric):
self.labels.append(truth)
for result in eval_results:
if isinstance(truth, str):
if isinstance(result, list):
result = result[0]
assert isinstance(result, str), 'both truth and pred are str'
self.preds.append(remove_space_between_chinese_chars(result))
else:
self.preds.append(result)

52
modelscope/models/multi_modal/ofa_for_all_tasks.py
View File

@@ -105,6 +105,8 @@ class OfaForAllTasks(TorchModel):
}
if hasattr(self.cfg.model, 'beam_search'):
sg_args.update(self.cfg.model.beam_search)
self.num_return_sequences = self.cfg.model.get('num_return_sequences',
1)
if len(self.ans2label_dict) > 0:
self.constraint_trie = Trie(self.tokenizer.eos_token_id)
self.val_ans_l = []
@@ -140,15 +142,14 @@ class OfaForAllTasks(TorchModel):
return self.inference(input)
def inference(self, input: Dict[str, Any]) -> Dict[str, Any]:
assert self.generator.beam_size >= self.num_return_sequences, \
'beam search can only return beam size sentences'
if self.ans2label_dict and self.gen_type == 'generation':
assert self.generator.beam_size <= len(self.ans2label_dict), \
'beam search will not work properly.'
ret = self.task_inference_mapping[self.cfg.task](input)
if 'samples' in input:
ret['samples'] = input['samples']
for key in [
OutputKeys.CAPTION, OutputKeys.TEXT, OutputKeys.BOXES,
OutputKeys.LABELS, OutputKeys.SCORES
]:
if key not in ret:
ret[key] = None
return ret
def postprocess(self, input: Dict[str, Any], **kwargs) -> Dict[str, Any]:
@@ -157,7 +158,8 @@ class OfaForAllTasks(TorchModel):
result_l = list()
for cap in caption:
if self.language == 'en':
result_l.append(cap.translate(self.transtab).strip())
result_l.append(
[c.translate(self.transtab).strip() for c in cap])
else:
result_l.append(cap)
input[OutputKeys.CAPTION] = result_l
@@ -166,8 +168,18 @@ class OfaForAllTasks(TorchModel):
] and self.cfg.task != Tasks.visual_grounding:
ret_l = list()
for text in input[OFA_TASK_KEY_MAPPING[self.cfg.task]]:
ret_l.append(self.detokenizer(text))
ret_l.append([self.detokenizer(t) for t in text])
input[OFA_TASK_KEY_MAPPING[self.cfg.task]] = ret_l
for key in [
OutputKeys.CAPTION, OutputKeys.TEXT, OutputKeys.BOXES,
OutputKeys.LABELS, OutputKeys.SCORES
]:
if key not in input:
input[key] = None
else:
if (len(input[key]) == 1 and isinstance(input[key], list)) \
and self.cfg.task != Tasks.visual_grounding:
input[key] = input[key][0]
return input
def _text_gen_inference(self, input):
@@ -175,23 +187,25 @@ class OfaForAllTasks(TorchModel):
input,
prefix_tokens=input.get(
'prefix_tokens', None))
gen_l = list()
results = list()
for idx, gen_out in enumerate(gen_outputs):
if len(gen_out) > 0:
decode_tokens = gen_out[0]['tokens']
gen_token_l = []
for beam_gen_out in gen_out[:self.num_return_sequences]:
decode_tokens = beam_gen_out['tokens']
if 'prefix_tokens' in input:
prefix_len = input['prefix_tokens'][idx].ne(
self.pad_item.to(self.model.device)).sum()
decode_tokens = decode_tokens[prefix_len:]
gen_l.append(decode_tokens)
else:
gen_l.append('')
result = self.tokenizer.batch_decode(gen_l, skip_special_tokens=True)
result = [item.strip() for item in result]
gen_token_l.append(decode_tokens)
result = self.tokenizer.batch_decode(
gen_token_l, skip_special_tokens=True)
result = [item.strip() for item in result]
result.extend([''] * (self.num_return_sequences - len(result)))
results.append(result)
# text generation tasks have no score
ret = {OFA_TASK_KEY_MAPPING[self.cfg.task]: result}
if self.cfg.task.endswith('classification'):
ret[OutputKeys.SCORES] = [1.0] * len(result)
ret = {OFA_TASK_KEY_MAPPING[self.cfg.task]: results}
if self.ans2label_dict:
ret[OutputKeys.SCORES] = [[1.0]] * len(results)
return ret
def _visual_grounding_inference(self, input):

19
modelscope/pipelines/base.py
View File

@@ -273,14 +273,17 @@ class Pipeline(ABC):
**forward_params)
else:
batched_out = self.forward(batched_input, **forward_params)
for batch_idx in range(real_batch_size):
out = {}
for k, element in batched_out.items():
if element is not None:
out[k] = element[batch_idx]
out = self.postprocess(out, **postprocess_params)
self._check_output(out)
output_list.append(out)
if real_batch_size == 1:
output_list.append(batched_out)
else:
for batch_idx in range(real_batch_size):
out = {}
for k, element in batched_out.items():
if element is not None:
out[k] = element[batch_idx]
out = self.postprocess(out, **postprocess_params)
self._check_output(out)
output_list.append(out)
return output_list

12
modelscope/pipelines/cv/image_classification_pipeline.py
View File

@@ -11,6 +11,7 @@ from modelscope.models.multi_modal import OfaForAllTasks
from modelscope.outputs import OutputKeys
from modelscope.pipelines.base import Input, Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import OfaPreprocessor, Preprocessor, load_image
from modelscope.utils.constant import Tasks
from modelscope.utils.device import get_device
@@ -35,6 +36,17 @@ class ImageClassificationPipeline(Pipeline):
if preprocessor is None and isinstance(self.model, OfaForAllTasks):
self.preprocessor = OfaPreprocessor(model_dir=self.model.model_dir)
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
return batch_process(self.model, data)
else:
return super(ImageClassificationPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():
return super().forward(inputs, **forward_params)
def postprocess(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
return inputs

8
modelscope/pipelines/multi_modal/asr_pipeline.py
View File

@@ -5,9 +5,9 @@ import torch
from modelscope.metainfo import Pipelines
from modelscope.models.multi_modal import MPlugForAllTasks, OfaForAllTasks
from modelscope.outputs import OutputKeys
from modelscope.pipelines.base import Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import (MPlugPreprocessor, OfaPreprocessor,
Preprocessor)
from modelscope.utils.constant import Tasks
@@ -45,6 +45,12 @@ class AutomaticSpeechRecognitionPipeline(Pipeline):
preprocessor = MPlugPreprocessor(pipe_model.model_dir)
super().__init__(model=pipe_model, preprocessor=preprocessor, **kwargs)
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
return batch_process(self.model, data)
else:
return super(AutomaticSpeechRecognitionPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():

16
modelscope/pipelines/multi_modal/image_captioning_pipeline.py
View File

@@ -5,9 +5,9 @@ import torch
from modelscope.metainfo import Pipelines
from modelscope.models.multi_modal import MPlugForAllTasks, OfaForAllTasks
from modelscope.outputs import OutputKeys
from modelscope.pipelines.base import Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import (MPlugPreprocessor, OfaPreprocessor,
Preprocessor)
from modelscope.utils.constant import Tasks
@@ -39,17 +39,7 @@ class ImageCaptioningPipeline(Pipeline):
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
# collate batch data due to the nested data structure
if isinstance(data, list):
batch_data = {}
batch_data['nsentences'] = len(data)
batch_data['samples'] = [d['samples'][0] for d in data]
batch_data['net_input'] = {}
for k in data[0]['net_input'].keys():
batch_data['net_input'][k] = torch.cat(
[d['net_input'][k] for d in data])
return batch_data
return batch_process(self.model, data)
elif isinstance(self.model, MPlugForAllTasks):
from transformers.tokenization_utils_base import BatchEncoding
batch_data = dict(train=data[0]['train'])
@@ -60,7 +50,7 @@ class ImageCaptioningPipeline(Pipeline):
batch_data['question'] = BatchEncoding(question)
return batch_data
else:
return super()._collate_batch(data)
return super(ImageCaptioningPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:

8
modelscope/pipelines/multi_modal/ocr_recognition_pipeline.py
View File

@@ -5,9 +5,9 @@ import torch
from modelscope.metainfo import Pipelines
from modelscope.models.multi_modal import OfaForAllTasks
from modelscope.outputs import OutputKeys
from modelscope.pipelines.base import Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import OfaPreprocessor, Preprocessor
from modelscope.utils.constant import Tasks
from modelscope.utils.logger import get_logger
@@ -34,6 +34,12 @@ class OcrRecognitionPipeline(Pipeline):
if isinstance(self.model, OfaForAllTasks):
self.preprocessor = OfaPreprocessor(self.model.model_dir)
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
return batch_process(self.model, data)
else:
return super(OcrRecognitionPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():

14
modelscope/pipelines/multi_modal/visual_entailment_pipeline.py
View File

@@ -1,10 +1,13 @@
# Copyright (c) Alibaba, Inc. and its affiliates.
from typing import Any, Dict, Optional, Union
import torch
from modelscope.metainfo import Pipelines
from modelscope.models.multi_modal import OfaForAllTasks
from modelscope.pipelines.base import Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import OfaPreprocessor, Preprocessor
from modelscope.utils.constant import Tasks
from modelscope.utils.logger import get_logger
@@ -30,5 +33,16 @@ class VisualEntailmentPipeline(Pipeline):
if preprocessor is None and isinstance(self.model, OfaForAllTasks):
self.preprocessor = OfaPreprocessor(model_dir=self.model.model_dir)
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
return batch_process(self.model, data)
else:
return super(VisualEntailmentPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():
return super().forward(inputs, **forward_params)
def postprocess(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
return inputs

14
modelscope/pipelines/multi_modal/visual_grounding_pipeline.py
View File

@@ -1,10 +1,13 @@
# Copyright (c) Alibaba, Inc. and its affiliates.
from typing import Any, Dict, Optional, Union
import torch
from modelscope.metainfo import Pipelines
from modelscope.models.multi_modal import OfaForAllTasks
from modelscope.pipelines.base import Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import OfaPreprocessor, Preprocessor
from modelscope.utils.constant import Tasks
from modelscope.utils.logger import get_logger
@@ -30,5 +33,16 @@ class VisualGroundingPipeline(Pipeline):
if preprocessor is None and isinstance(self.model, OfaForAllTasks):
self.preprocessor = OfaPreprocessor(model_dir=self.model.model_dir)
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
return batch_process(self.model, data)
else:
return super(VisualGroundingPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():
return super().forward(inputs, **forward_params)
def postprocess(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
return inputs

8
modelscope/pipelines/multi_modal/visual_question_answering_pipeline.py
View File

@@ -6,9 +6,9 @@ import torch
from modelscope.metainfo import Pipelines
from modelscope.models import Model
from modelscope.models.multi_modal import MPlugForAllTasks, OfaForAllTasks
from modelscope.outputs import OutputKeys
from modelscope.pipelines.base import Pipeline, Tensor
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import (MPlugPreprocessor, OfaPreprocessor,
Preprocessor)
from modelscope.utils.constant import Tasks
@@ -39,6 +39,12 @@ class VisualQuestionAnsweringPipeline(Pipeline):
self.preprocessor = MPlugPreprocessor(self.model.model_dir)
self.model.eval()
def _batch(self, data):
if isinstance(self.model, OfaForAllTasks):
return batch_process(self.model, data)
else:
return super(VisualQuestionAnsweringPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():

14
modelscope/pipelines/nlp/summarization_pipeline.py
View File

@@ -1,9 +1,12 @@
# Copyright (c) Alibaba, Inc. and its affiliates.
from typing import Any, Dict, Optional, Union
import torch
from modelscope.metainfo import Pipelines, Preprocessors
from modelscope.pipelines.base import Model, Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import Preprocessor
from modelscope.utils.constant import Fields, Tasks
from modelscope.utils.logger import get_logger
@@ -48,5 +51,16 @@ class SummarizationPipeline(Pipeline):
self.preprocessor = Preprocessor.from_pretrained(
self.model.model_dir, **kwargs)
def _batch(self, data):
if self.model.__class__.__name__ == 'OfaForAllTasks':
return batch_process(self.model, data)
else:
return super(SummarizationPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
with torch.no_grad():
return super().forward(inputs, **forward_params)
def postprocess(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
return inputs

11
modelscope/pipelines/nlp/text_classification_pipeline.py
View File

@@ -2,12 +2,14 @@
from typing import Any, Dict, Union
import numpy as np
import torch
from modelscope.metainfo import Pipelines, Preprocessors
from modelscope.models.base import Model
from modelscope.outputs import OutputKeys, TextClassificationModelOutput
from modelscope.pipelines.base import Pipeline
from modelscope.pipelines.builder import PIPELINES
from modelscope.pipelines.util import batch_process
from modelscope.preprocessors import Preprocessor
from modelscope.utils.constant import Fields, Tasks
from modelscope.utils.logger import get_logger
@@ -83,10 +85,17 @@ class TextClassificationPipeline(Pipeline):
if hasattr(self.preprocessor, 'id2label'):
self.id2label = self.preprocessor.id2label
def _batch(self, data):
if self.model.__class__.__name__ == 'OfaForAllTasks':
return batch_process(self.model, data)
else:
return super(TextClassificationPipeline, self)._batch(data)
def forward(self, inputs: Dict[str, Any],
**forward_params) -> Dict[str, Any]:
if self.model.__class__.__name__ == 'OfaForAllTasks':
return super().forward(inputs, **forward_params)
with torch.no_grad():
return super().forward(inputs, **forward_params)
return self.model(**inputs, **forward_params)
def postprocess(self,

24
modelscope/pipelines/util.py
View File

@@ -2,6 +2,8 @@
import os.path as osp
from typing import List, Optional, Union
import torch
from modelscope.hub.api import HubApi
from modelscope.hub.file_download import model_file_download
from modelscope.utils.config import Config
@@ -81,3 +83,25 @@ def is_model(path: Union[str, List]):
)
return all_true
def batch_process(model, data):
if model.__class__.__name__ == 'OfaForAllTasks':
# collate batch data due to the nested data structure
assert isinstance(data, list)
batch_data = {
'nsentences': len(data),
'samples': [d['samples'][0] for d in data],
'net_input': {}
}
for k in data[0]['net_input'].keys():
batch_data['net_input'][k] = torch.cat(
[d['net_input'][k] for d in data])
if 'w_resize_ratios' in data[0]:
batch_data['w_resize_ratios'] = torch.cat(
[d['w_resize_ratios'] for d in data])
if 'h_resize_ratios' in data[0]:
batch_data['h_resize_ratios'] = torch.cat(
[d['h_resize_ratios'] for d in data])
return batch_data

3
modelscope/preprocessors/ofa/image_classification.py
View File

@@ -112,7 +112,8 @@ class OfaImageClassificationPreprocessor(OfaBasePreprocessor):
sample = {
'source': inputs,
'patch_image': patch_image,
'patch_mask': torch.tensor([True])
'patch_mask': torch.tensor([True]),
'decoder_prompt': self.bos_item,
}
if 'text' in self.column_map and self.column_map['text'] in data:
sample['label'] = data[self.column_map['text']]

3
modelscope/preprocessors/ofa/text_classification.py
View File

@@ -68,13 +68,16 @@ class OfaTextClassificationPreprocessor(OfaBasePreprocessor):
instruction_itm = self._build_instruction(data)
if self.prompt_type == 'none':
prefix_token = []
decoder_prompt = self.bos_item
elif self.prompt_type == 'prev_output':
prefix_token = instruction_itm[:-1] # remove eos
decoder_prompt = instruction_itm[:-1]
else:
raise NotImplementedError
sample = {
'source': instruction_itm,
'prefix_token': prefix_token,
'decoder_prompt': decoder_prompt,
}
if 'label' in data:
sample['label'] = self.label2ans[data['label']]

5
modelscope/preprocessors/ofa/visual_entailment.py
View File

@@ -101,10 +101,10 @@ class OfaVisualEntailmentPreprocessor(OfaBasePreprocessor):
text = prompt.format(caption, hypothesis)
inputs = self.tokenize_text(text)
if self.prompt_type == 'none':
prefix_token = []
decoder_prompt = self.bos_item
elif self.prompt_type == 'src':
decoder_prompt = inputs
elif self.prompt_type == 'prev_output':
prefix_token = inputs[:-1] # remove eos
decoder_prompt = inputs[:-1]
else:
raise NotImplementedError
@@ -112,6 +112,7 @@ class OfaVisualEntailmentPreprocessor(OfaBasePreprocessor):
'source': inputs,
'patch_image': patch_image,
'patch_mask': torch.tensor([True]),
'prefix_token': prefix_token,
'decoder_prompt': decoder_prompt,
}
if 'relation' in self.column_map and self.column_map[

64
tests/pipelines/test_ofa_tasks.py
View File

@@ -45,15 +45,16 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
result = img_captioning('data/test/images/image_captioning.png')
print(result[OutputKeys.CAPTION])
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
def test_run_with_image_captioning_batch(self):
img_captioning = pipeline(
Tasks.image_captioning,
model='damo/ofa_image-caption_coco_large_en')
img_captioning.model.num_return_sequences = 2
result = img_captioning('data/test/images/image_captioning.png')
print(result[OutputKeys.CAPTION])
# test batch infer
img_captioning.model.num_return_sequences = 1
results = img_captioning(
[{
'image': 'data/test/images/image_captioning.png'
} for _ in range(6)],
} for _ in range(3)],
batch_size=2)
for r in results:
print(r[OutputKeys.CAPTION])
@@ -65,6 +66,12 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
model='damo/ofa_ocr-recognition_scene_base_zh')
result = ocr_recognize('data/test/images/image_ocr_recognition.jpg')
print(result[OutputKeys.TEXT])
# test batch infer
results = ocr_recognize(
['data/test/images/image_ocr_recognition.jpg' for _ in range(3)],
batch_size=2)
for r in results:
print(r[OutputKeys.TEXT])
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
def test_run_with_image_classification_with_model(self):
@@ -84,6 +91,12 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
result = ofa_pipe(image)
print(result)
# test batch infer
image = ['data/test/images/image_classification.png' for _ in range(3)]
results = ofa_pipe(image, batch_size=2)
for r in results:
print(r[OutputKeys.LABELS], r[OutputKeys.SCORES])
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
def test_run_with_summarization_with_model(self):
model = Model.from_pretrained(
@@ -104,12 +117,23 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
model='damo/ofa_summarization_gigaword_large_en')
text = 'five-time world champion michelle kwan withdrew' + \
'from the #### us figure skating championships on wednesday ,' + \
' but will petition us skating officials for the chance to ' +\
' but will petition us skating officials for the chance to ' + \
'compete at the #### turin olympics .'
input = {'text': text}
result = ofa_pipe(input)
print(result)
# test for return multiple sequences
ofa_pipe.model.num_return_sequences = 2
result = ofa_pipe(input)
print(result)
# test batch infer
ofa_pipe.model.num_return_sequences = 1
input = [{'text': text} for _ in range(3)]
results = ofa_pipe(input, batch_size=2)
for r in results:
print(r[OutputKeys.TEXT])
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
def test_run_with_text_classification_with_model(self):
model = Model.from_pretrained(
@@ -130,6 +154,11 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
text2 = 'A member of my team will execute your orders with immense precision.'
result = ofa_pipe((text, text2))
print(result)
# test batch infer
inputs = [(text, text2) for _ in range(3)]
results = ofa_pipe(inputs, batch_size=2)
for r in results:
print(r[OutputKeys.LABELS], r[OutputKeys.SCORES])
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
def test_run_with_visual_entailment_with_model(self):
@@ -152,8 +181,13 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
input = {'image': image, 'text': text}
result = ofa_pipe(input)
print(result)
# test batch infer
input = [{'image': image, 'text': text} for _ in range(3)]
results = ofa_pipe(input, batch_size=2)
for r in results:
print(r[OutputKeys.LABELS], r[OutputKeys.SCORES])
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
def test_run_with_visual_grounding_with_model(self):
model = Model.from_pretrained(
'damo/ofa_visual-grounding_refcoco_large_en')
@@ -182,6 +216,9 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
image_name = image.split('/')[-2]
self.save_img(image, result[OutputKeys.BOXES][0],
osp.join('large_en_name_' + image_name + '.png'))
# test batch infer
result = ofa_pipe([input for _ in range(3)], batch_size=2)
print(result)
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
def test_run_with_visual_grounding_zh_with_name(self):
@@ -217,6 +254,10 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
result = ofa_pipe(input)
print(result)
# test batch infer
result = ofa_pipe([input for _ in range(3)], batch_size=2)
print(result)
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
def test_run_with_image_captioning_distilled_with_model(self):
model = Model.from_pretrained(
@@ -230,6 +271,9 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
result = img_captioning(image)
print(result[OutputKeys.CAPTION])
# test batch infer
print(img_captioning([image for _ in range(3)], batch_size=2))
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
def test_run_with_visual_entailment_distilled_model_with_name(self):
ofa_pipe = pipeline(
@@ -280,6 +324,10 @@ class OfaTasksTest(unittest.TestCase, DemoCompatibilityCheck):
example = {'wav': 'data/test/audios/asr_example_ofa.wav'}
result = ofa_pipe(example)
print(result[OutputKeys.TEXT])
# test batch infer
result = ofa_pipe([example for _ in range(3)], batch_size=2)
for r in result:
print(r[OutputKeys.TEXT])
@unittest.skip('demo compatibility test is only enabled on a needed-basis')
def test_demo_compatibility(self):

2
tests/trainers/test_ofa_trainer.py
View File

@@ -37,7 +37,7 @@ class TestOfaTrainer(unittest.TestCase):
'train': {'work_dir': 'work/ckpts/recognition',
# 'launcher': 'pytorch',
'max_epochs': 1,
'use_fp16': True,
'use_fp16': False,
'dataloader': {'batch_size_per_gpu': 4, 'workers_per_gpu': 0},
'lr_scheduler': {'name': 'polynomial_decay',
'warmup_proportion': 0.01,