def __init__(self, config: Union[Dict[str, Any], Callable[..., None]],
**kwargs):
super().__init__()
if isinstance(config, Dict):
config = builder.build_config(config)
self.config = config
self.weights_name = "model_pt.bin"
def setup_class(cls):
cls.config_file_path = "/workspace/models/nlp/chinese_wwm_ext/bert_config.json"
cls.tf_checkpoint_path = "/workspace/models/nlp/chinese_wwm_ext/bert_model.ckpt"
cls.huggingface_model_path = "/workspace/models/nlp/chinese_wwm_ext"
cls.model_path = "/workspace/models/nlp/chinese_wwm_ext/bert_model_ms.ckpt"
model_cfg = dict(
type="MSBertForPreTraining",
config=dict(type="ConfigBase", json_file=cls.config_file_path),
)
cls.config = build_config(model_cfg["config"])
cls.model_tf = build_ms_models(model_cfg)
cls.model_hf = build_ms_models(model_cfg)
cls.model_base = transformers.BertModel.from_pretrained(
cls.huggingface_model_path, config=cls.config_file_path)
cls.model_base.eval()
# cls.model_base_mlm = transformers.BertForPreTraining.from_pretrained(
# cls.huggingface_model_path, config=cls.config_file_path
# )
# cls.model_base_mlm.eval()
model_cfg.update({"model_path": cls.model_path})
cls.model = build_ms_models(model_cfg)
cls.batch_size = 4
cls.seq_length = 10
cls.tokens_tensor = {
"input_ids":
ops.uniform(
shape=(cls.batch_size, cls.seq_length),
minval=Tensor(1, mindspore.int32),
maxval=Tensor(100, mindspore.int32),
dtype=mindspore.int32,
),
"attention_mask":
ops.uniform(
shape=(cls.batch_size, cls.seq_length),
minval=Tensor(0, mindspore.int32),
maxval=Tensor(2, mindspore.int32),
dtype=mindspore.int32,
),
"token_type_ids":
ops.uniform(
shape=(cls.batch_size, cls.seq_length),
minval=Tensor(0, mindspore.int32),
maxval=Tensor(2, mindspore.int32),
dtype=mindspore.int32,
),
"position_ids":
ops.uniform(
shape=(cls.batch_size, cls.seq_length),
minval=Tensor(0, mindspore.int32),
maxval=Tensor(cls.seq_length, mindspore.int32),
dtype=mindspore.int32,
),
}
def setup_class(cls):
cls.config_file_path = "/workspace/models/nlp/chinese_wwm_ext/bert_config.json"
cls.tf_checkpoint_path = "/workspace/models/nlp/chinese_wwm_ext/bert_model.ckpt"
cls.huggingface_model_path = "/workspace/models/nlp/chinese_wwm_ext"
cls.model_path = "/workspace/models/nlp/chinese_wwm_ext/bert_model_pd.bin"
model_cfg = dict(
type="PDBertForPreTraining",
config=dict(type="ConfigBase", json_file=cls.config_file_path),
)
cls.config = build_config(model_cfg["config"])
cls.model_tf = build_pd_models(model_cfg)
cls.model_hf = build_pd_models(model_cfg)
cls.model_base = transformers.BertModel.from_pretrained(
cls.huggingface_model_path)
cls.model_base.eval()
cls.model_base_mlm = transformers.BertForPreTraining.from_pretrained(
cls.huggingface_model_path)
cls.model_base_mlm.eval()
model_cfg.update({"model_path": cls.model_path})
cls.model = build_pd_models(model_cfg)
cls.model.eval()
cls.batch_size = 4
cls.seq_length = 10
cls.tokens_tensor = {
"input_ids":
paddle.randint(
low=1,
high=100,
shape=(cls.batch_size, cls.seq_length),
dtype=paddle.int64),
"attention_mask":
paddle.randint(
low=0,
high=2,
shape=(cls.batch_size, cls.seq_length),
dtype=paddle.int64),
"token_type_ids":
paddle.randint(
low=0,
high=2,
shape=(cls.batch_size, cls.seq_length),
dtype=paddle.int64),
"position_ids":
paddle.randint(
low=0,
high=cls.seq_length,
shape=(cls.batch_size, cls.seq_length),
dtype=paddle.int64,
),
}
def setup_class(cls):
cls.config_file_path = "/workspace/models/nlp/uncased_L-12_H-768_A-12/bert_config.json"
cls.tf_checkpoint_path = "/workspace/models/nlp/uncased_L-12_H-768_A-12/bert_model.ckpt"
cls.huggingface_model_path = "/workspace/models/nlp/uncased_L-12_H-768_A-12"
cls.model_path = "/workspace/models/nlp/uncased_L-12_H-768_A-12/bert_model_pt.bin"
model_cfg = dict(
type="BertForPreTraining",
config=dict(type="ConfigBase", json_file=cls.config_file_path),
)
cls.config = build_config(model_cfg["config"])
cls.model_tf = build_torch_models(model_cfg)
cls.model_hf = build_torch_models(model_cfg)
cls.model_base = transformers.BertModel.from_pretrained(
cls.huggingface_model_path, return_dict=True)
cls.model_base.eval()
cls.model_base_mlm = transformers.BertForPreTraining.from_pretrained(
cls.huggingface_model_path, return_dict=True)
cls.model_base_mlm.eval()
model_cfg.update({"model_path": cls.model_path})
cls.model = build_torch_models(model_cfg)
cls.model.eval()
cls.batch_size = 4
cls.seq_length = 10
cls.tokens_tensor = {
"input_ids":
torch.randint(low=1,
high=100,
size=(cls.batch_size, cls.seq_length),
dtype=torch.long),
"attention_mask":
torch.randint(low=0,
high=1,
size=(cls.batch_size, cls.seq_length),
dtype=torch.long),
"token_type_ids":
torch.randint(low=0,
high=1,
size=(cls.batch_size, cls.seq_length),
dtype=torch.long),
"position_ids":
torch.randint(low=0,
high=cls.batch_size,
size=(cls.batch_size, cls.seq_length),
dtype=torch.long),
}