Tianlei Wu
3ec3e9f705
Add t-test to compare experiments in GPT-2 mixed precision conversion ( #9042 )
...
* Add t-test to compare two experiments
* Ranking based on pair-wise T-test results and a custom scoring function
2021-09-14 12:40:25 -07:00
Tianlei Wu
91f05f387a
Update embed layer norm fusion to work with transformers v4.9 ( #8914 )
2021-09-02 19:48:07 -07:00
Tianlei Wu
f6e7cd8050
remove existed dir for output external data ( #8887 )
2021-08-30 21:23:07 -07:00
Tianlei Wu
cb59f46e04
Add gpt2 mixed precision conversion and parity tools ( #8845 )
2021-08-26 15:34:45 -07:00
Tianlei Wu
41f1280fc9
Fix transformer optimizer ( #8392 )
...
* fix a few issues
2021-07-14 16:00:17 -07:00
Cecilia Liu
4fd9fef9ee
Support HuggingFace Models Converted From tf2onnx in Python Script ( #6985 )
...
Support tf2onnx huggingface models in python script
2021-03-17 15:33:57 -07:00
Tianlei Wu
f4acdb2ecd
Update transformers benchmark for transformers 4.3.* and ORT 1.7 ( #6796 )
...
* update benchmark for transformers 4.* and ORT 1.7
* Fix gpt2 onnx conversion for transformers 4.3.*. Add a check of transformer version >= 3.1.
* remove code related to openmp
* update pretrain model list: keep representitive models only
2021-02-24 12:52:35 -08:00
Tianlei Wu
ac5ca2bbe0
fix data_ptr assertion error for past_sequence_length=0 in GPT-2 ( #6284 )
...
fix io binding crash for past_sequence_length=0
2021-01-07 23:43:50 -08:00
Tianlei Wu
b80e8ce6a5
rename past to past_key_values for GPT-2 ( #6269 )
...
rename past to past_key_values for transformers 4.*
2021-01-07 11:12:04 -08:00
Tianlei Wu
094384781e
Add --use_external_data_format in convert_to_onnx.py ( #5393 )
2020-10-07 09:42:02 -07:00
Tianlei Wu
7511021e0e
Save Gpt2 test data ( #5132 )
...
(1) Save gpt2 test data during test generation.
(2) Use torch fp32 model as baseline when onnx model is fp16.
(3) Refine logic to compose onnx model path
2020-09-11 14:31:49 -07:00
Tianlei Wu
c5d4ae0401
Add transformers tools to python package ( #5090 )
...
* Add transformers to onnxruntime python package
2020-09-10 15:42:15 -07:00
Tianlei Wu
268d2283c0
Export GPT-2 ONNX model without postion_ids and attention_mask inputs ( #4852 )
...
* Export GPT-2 ONNX model without postion_ids and attention_mask inputs
* allow benchmark_gpt2 on user's model
* refactor: get_dummy_inputs returns a data class.
2020-08-24 13:05:25 -07:00
Tianlei Wu
316d1a9e69
Update benchmark for large model or model name with non-alphanumeric. ( #4743 )
...
* Export model > 2GB using external data format
2020-08-10 12:58:01 -07:00
Tianlei Wu
3588c5b545
Add GPT-2 test generation to convert_to_onnx.py ( #4670 )
...
* add gpt2 tester
* add an option to include output latency.
2020-07-30 21:03:53 -07:00
Tianlei Wu
ea87c0d028
Update Transformer Optimizer documents ( #4591 )
...
(1) Add bert-base-cased and gpt2 benchmark results on V100
(2) Update list of supported models.
(3) Add comments to gpt2_helper.
(4) Use IO Binding in test parity by default.
2020-07-23 08:38:39 -07:00
Tianlei Wu
b42fe49f56
Add a script to convert gpt2 to onnx ( #4541 )
...
* add a script convert_to_onnx.py and refactor benchmark
2020-07-17 23:14:42 -07:00
