2020-05-14 23:35:52 +03:00
|
|
|
import unittest
|
|
|
|
from os.path import dirname, exists
|
2020-07-29 14:21:29 +03:00
|
|
|
from pathlib import Path
|
2020-05-14 23:35:52 +03:00
|
|
|
from shutil import rmtree
|
2020-06-01 17:12:48 +03:00
|
|
|
from tempfile import NamedTemporaryFile, TemporaryDirectory
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
from transformers import BertConfig, BertTokenizerFast, FeatureExtractionPipeline
|
2020-07-29 14:21:29 +03:00
|
|
|
from transformers.convert_graph_to_onnx import (
|
|
|
|
convert,
|
|
|
|
ensure_valid_input,
|
|
|
|
generate_identified_filename,
|
|
|
|
infer_shapes,
|
|
|
|
quantize,
|
|
|
|
)
|
2020-07-01 17:31:17 +03:00
|
|
|
from transformers.testing_utils import require_tf, require_torch, slow
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
|
|
|
|
class FuncContiguousArgs:
|
|
|
|
def forward(self, input_ids, token_type_ids, attention_mask):
|
|
|
|
return None
|
|
|
|
|
|
|
|
|
|
|
|
class FuncNonContiguousArgs:
|
|
|
|
def forward(self, input_ids, some_other_args, token_type_ids, attention_mask):
|
|
|
|
return None
|
|
|
|
|
|
|
|
|
|
|
|
class OnnxExportTestCase(unittest.TestCase):
|
|
|
|
MODEL_TO_TEST = ["bert-base-cased", "gpt2", "roberta-base"]
|
|
|
|
|
|
|
|
@require_tf
|
2020-05-18 16:24:41 +03:00
|
|
|
@slow
|
2020-05-14 23:35:52 +03:00
|
|
|
def test_export_tensorflow(self):
|
|
|
|
for model in OnnxExportTestCase.MODEL_TO_TEST:
|
2020-07-29 14:21:29 +03:00
|
|
|
self._test_export(model, "tf", 12)
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
@require_torch
|
2020-05-18 16:24:41 +03:00
|
|
|
@slow
|
2020-05-14 23:35:52 +03:00
|
|
|
def test_export_pytorch(self):
|
|
|
|
for model in OnnxExportTestCase.MODEL_TO_TEST:
|
2020-07-29 14:21:29 +03:00
|
|
|
self._test_export(model, "pt", 12)
|
2020-05-14 23:35:52 +03:00
|
|
|
|
2020-06-01 17:12:48 +03:00
|
|
|
@require_torch
|
|
|
|
@slow
|
|
|
|
def test_export_custom_bert_model(self):
|
|
|
|
from transformers import BertModel
|
|
|
|
|
|
|
|
vocab = ["[UNK]", "[SEP]", "[CLS]", "[PAD]", "[MASK]", "some", "other", "words"]
|
|
|
|
with NamedTemporaryFile(mode="w+t") as vocab_file:
|
|
|
|
vocab_file.write("\n".join(vocab))
|
|
|
|
vocab_file.flush()
|
|
|
|
tokenizer = BertTokenizerFast(vocab_file.name)
|
|
|
|
|
|
|
|
with TemporaryDirectory() as bert_save_dir:
|
|
|
|
model = BertModel(BertConfig(vocab_size=len(vocab)))
|
|
|
|
model.save_pretrained(bert_save_dir)
|
2020-07-29 14:21:29 +03:00
|
|
|
self._test_export(bert_save_dir, "pt", 12, tokenizer)
|
|
|
|
|
|
|
|
@require_tf
|
|
|
|
@slow
|
|
|
|
def test_quantize_tf(self):
|
|
|
|
for model in OnnxExportTestCase.MODEL_TO_TEST:
|
|
|
|
path = self._test_export(model, "tf", 12)
|
|
|
|
quantized_path = quantize(Path(path))
|
|
|
|
|
|
|
|
# Ensure the actual quantized model is not bigger than the original one
|
|
|
|
if quantized_path.stat().st_size >= Path(path).stat().st_size:
|
|
|
|
self.fail("Quantized model is bigger than initial ONNX model")
|
|
|
|
|
|
|
|
@require_torch
|
|
|
|
@slow
|
|
|
|
def test_quantize_pytorch(self):
|
|
|
|
for model in OnnxExportTestCase.MODEL_TO_TEST:
|
|
|
|
path = self._test_export(model, "pt", 12)
|
|
|
|
quantized_path = quantize(Path(path))
|
|
|
|
|
|
|
|
# Ensure the actual quantized model is not bigger than the original one
|
|
|
|
if quantized_path.stat().st_size >= Path(path).stat().st_size:
|
|
|
|
self.fail("Quantized model is bigger than initial ONNX model")
|
2020-06-01 17:12:48 +03:00
|
|
|
|
|
|
|
def _test_export(self, model, framework, opset, tokenizer=None):
|
2020-05-14 23:35:52 +03:00
|
|
|
try:
|
|
|
|
# Compute path
|
2020-06-01 17:12:48 +03:00
|
|
|
with TemporaryDirectory() as tempdir:
|
|
|
|
path = tempdir + "/model.onnx"
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
# Remove folder if exists
|
|
|
|
if exists(dirname(path)):
|
|
|
|
rmtree(dirname(path))
|
|
|
|
|
2020-06-01 17:12:48 +03:00
|
|
|
# Export
|
|
|
|
convert(framework, model, path, opset, tokenizer)
|
2020-07-29 14:21:29 +03:00
|
|
|
|
|
|
|
return path
|
2020-05-14 23:35:52 +03:00
|
|
|
except Exception as e:
|
|
|
|
self.fail(e)
|
|
|
|
|
|
|
|
@require_torch
|
|
|
|
def test_infer_dynamic_axis_pytorch(self):
|
|
|
|
"""
|
|
|
|
Validate the dynamic axis generated for each parameters are correct
|
|
|
|
"""
|
|
|
|
from transformers import BertModel
|
|
|
|
|
|
|
|
model = BertModel(BertConfig.from_pretrained("bert-base-cased"))
|
|
|
|
tokenizer = BertTokenizerFast.from_pretrained("bert-base-cased")
|
|
|
|
self._test_infer_dynamic_axis(model, tokenizer, "pt")
|
|
|
|
|
|
|
|
@require_tf
|
|
|
|
def test_infer_dynamic_axis_tf(self):
|
|
|
|
"""
|
|
|
|
Validate the dynamic axis generated for each parameters are correct
|
|
|
|
"""
|
|
|
|
from transformers import TFBertModel
|
|
|
|
|
|
|
|
model = TFBertModel(BertConfig.from_pretrained("bert-base-cased"))
|
|
|
|
tokenizer = BertTokenizerFast.from_pretrained("bert-base-cased")
|
|
|
|
self._test_infer_dynamic_axis(model, tokenizer, "tf")
|
|
|
|
|
|
|
|
def _test_infer_dynamic_axis(self, model, tokenizer, framework):
|
|
|
|
nlp = FeatureExtractionPipeline(model, tokenizer)
|
|
|
|
|
|
|
|
variable_names = ["input_ids", "token_type_ids", "attention_mask", "output_0", "output_1"]
|
|
|
|
input_vars, output_vars, shapes, tokens = infer_shapes(nlp, framework)
|
|
|
|
|
|
|
|
# Assert all variables are present
|
|
|
|
self.assertEqual(len(shapes), len(variable_names))
|
|
|
|
self.assertTrue(all([var_name in shapes for var_name in variable_names]))
|
|
|
|
self.assertSequenceEqual(variable_names[:3], input_vars)
|
|
|
|
self.assertSequenceEqual(variable_names[3:], output_vars)
|
|
|
|
|
|
|
|
# Assert inputs are {0: batch, 1: sequence}
|
|
|
|
for var_name in ["input_ids", "token_type_ids", "attention_mask"]:
|
|
|
|
self.assertDictEqual(shapes[var_name], {0: "batch", 1: "sequence"})
|
|
|
|
|
|
|
|
# Assert outputs are {0: batch, 1: sequence} and {0: batch}
|
|
|
|
self.assertDictEqual(shapes["output_0"], {0: "batch", 1: "sequence"})
|
|
|
|
self.assertDictEqual(shapes["output_1"], {0: "batch"})
|
|
|
|
|
|
|
|
def test_ensure_valid_input(self):
|
|
|
|
"""
|
|
|
|
Validate parameters are correctly exported
|
|
|
|
GPT2 has "past" parameter in the middle of input_ids, token_type_ids and attention_mask.
|
|
|
|
ONNX doesn't support export with a dictionary, only a tuple. Thus we need to ensure we remove
|
|
|
|
token_type_ids and attention_mask for now to not having a None tensor in the middle
|
|
|
|
"""
|
|
|
|
# All generated args are valid
|
|
|
|
input_names = ["input_ids", "attention_mask", "token_type_ids"]
|
|
|
|
tokens = {"input_ids": [1, 2, 3, 4], "attention_mask": [0, 0, 0, 0], "token_type_ids": [1, 1, 1, 1]}
|
2020-06-01 17:12:48 +03:00
|
|
|
ordered_input_names, inputs_args = ensure_valid_input(FuncContiguousArgs(), tokens, input_names)
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
# Should have exactly the same number of args (all are valid)
|
|
|
|
self.assertEqual(len(inputs_args), 3)
|
|
|
|
|
2020-06-01 17:12:48 +03:00
|
|
|
# Should have exactly the same input names
|
|
|
|
self.assertEqual(set(ordered_input_names), set(input_names))
|
|
|
|
|
2020-05-14 23:35:52 +03:00
|
|
|
# Parameter should be reordered according to their respective place in the function:
|
|
|
|
# (input_ids, token_type_ids, attention_mask)
|
|
|
|
self.assertEqual(inputs_args, (tokens["input_ids"], tokens["token_type_ids"], tokens["attention_mask"]))
|
|
|
|
|
|
|
|
# Generated args are interleaved with another args (for instance parameter "past" in GPT2)
|
2020-06-01 17:12:48 +03:00
|
|
|
ordered_input_names, inputs_args = ensure_valid_input(FuncNonContiguousArgs(), tokens, input_names)
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
# Should have exactly the one arg (all before the one not provided "some_other_args")
|
|
|
|
self.assertEqual(len(inputs_args), 1)
|
2020-06-01 17:12:48 +03:00
|
|
|
self.assertEqual(len(ordered_input_names), 1)
|
2020-05-14 23:35:52 +03:00
|
|
|
|
|
|
|
# Should have only "input_ids"
|
|
|
|
self.assertEqual(inputs_args[0], tokens["input_ids"])
|
2020-06-01 17:12:48 +03:00
|
|
|
self.assertEqual(ordered_input_names[0], "input_ids")
|
2020-07-29 14:21:29 +03:00
|
|
|
|
|
|
|
def test_generate_identified_name(self):
|
|
|
|
generated = generate_identified_filename(Path("/home/something/my_fake_model.onnx"), "-test")
|
|
|
|
self.assertEqual("/home/something/my_fake_model-test.onnx", generated.as_posix())
|