add infer predict function for PET #2690
Conversation
add infer predict function for PET
add infer predict function for PET
|
infer predict example model_dir is the path of infer model exist |
| from paddlenlp.data import Stack, Tuple, Pad | ||
| from paddlenlp.datasets import load_dataset | ||
| from paddlenlp.utils.log import logger | ||
| from functools import partial |
| tokenizer = ppnlp.transformers.ErnieTokenizer.from_pretrained('ernie-1.0') | ||
|
|
||
| batchify_test_fn = lambda samples, fn=Tuple( | ||
| Pad(axis=0, pad_val=tokenizer.pad_token_id), # src_ids |
There was a problem hiding this comment.
Pad这块可以设置一下dtype="int64", 在windows不会出现不兼容的问题
| parser.add_argument("--pattern_id", default=0, type=int, help="pattern id of pet") | ||
| args = parser.parse_args() | ||
|
|
||
|
|
| self.batch_size = batch_size | ||
|
|
||
| model_file = model_dir + "/inference.pdmodel" | ||
| params_file = model_dir + "/inference.pdiparams" |
There was a problem hiding this comment.
这块是不是可以os.path.join这种，在windows的目录层次是 \ 符号
| self.output_handle = self.predictor.get_output_handle( | ||
| self.predictor.get_output_names()[0]) | ||
|
|
||
| if args.benchmark: |
There was a problem hiding this comment.
benchmark这块的代码可以去掉，PET这块不需要benchmark监测
| parser.add_argument('--enable_mkldnn', default=False, type=eval, choices=[True, False], | ||
| help='Enable to use mkldnn to speed up when using cpu.') | ||
|
|
||
| parser.add_argument("--benchmark", type=eval, default=False, |
| batchify_fn=batchify_test_fn, | ||
| trans_fn=trans_test_func) | ||
| p = Predictor(args.model_dir) | ||
| y = p.predict(tokenizer, test_data_loader, label_norm_dict) |
There was a problem hiding this comment.
整体需要一个文档介绍一下怎么使用，例如TRT的安装，可以拷贝一下其他的文档；同时predict.py怎么使用
There was a problem hiding this comment.
add README.md for how to use predict.py
There was a problem hiding this comment.
add README.md for how to use predict.py
OK，Thank for your contribution！
There was a problem hiding this comment.
add README.md for how to use predict.py
OK，Thank for your contribution！
Thanks for your patience while i make these changes
PR types
OthersPR changes
Models
Description
add infer predict function for PET