def get_model(self):
self.model = TENER(tag_vocab=self.data_bundle.get_vocab('target'), embed=self.embed, num_layers=num_layers,
d_model=d_model, n_head=n_heads,
feedforward_dim=dim_feedforward, dropout=dropout,
after_norm=after_norm, attn_type=attn_type,
bi_embed=None,
fc_dropout=fc_dropout,
pos_embed=pos_embed,
scale=attn_type=='transformer')
#Thay model path trước khi chạy
model_path = './model/model.bin'
states = torch.load(model_path).state_dict()
self.model.load_state_dict(states)
return self.model, self.data
def make_model(model_id, dataset, config):
if model_id == config['model_ids']['tener']:
data_bundle, embed = get_data_loading_function(dataset,
config)(dataset, config)
return TENER(data_bundle=data_bundle,
config=config,
embed=embed,
num_layers=config['num_layers'],
d_model=config['n_heads'] * config['head_dims'],
n_head=config['n_heads'],
feedforward_dim=int(2 * config['n_heads'] *
config['head_dims']),
dropout=config['dropout'],
after_norm=config['after_norm'],
attn_type=config['attn_type'],
bi_embed=None,
fc_dropout=config['fc_dropout'],
pos_embed=config['pos_embed'],
scale=config['attn_type'] == 'transformer')
dev_data = list(data_bundle.get_dataset("dev"))
test_data = list(data_bundle.get_dataset("test"))
vocab_size = len(data_bundle.get_vocab('chars'))
feature_vocab_size = len(feature2id)
model = TENER(tag_vocab=data_bundle.get_vocab('target'),
embed=embed,
num_layers=num_layers,
d_model=d_model,
n_head=n_heads,
feedforward_dim=dim_feedforward,
dropout=trans_dropout,
after_norm=after_norm,
attn_type=attn_type,
bi_embed=None,
fc_dropout=fc_dropout,
pos_embed=pos_embed,
scale=attn_type == 'naive',
vocab_size=vocab_size,
feature_vocab_size=feature_vocab_size,
kv_attn_type=kv_attn_type,
memory_dropout=memory_dropout,
fusion_dropout=fusion_dropout,
fusion_type=fusion_type,
highway_layer=highway_layer)
optimizer = optim.Adam(model.parameters(), lr=lr, betas=(0.9, 0.99))
callbacks = []
clip_callback = GradientClipCallback(clip_type='value', clip_value=5)
evaluate_callback = EvaluateCallback(data=data_bundle.get_dataset('test'),
dropout=0,
word_dropout=0.02)
return data, embed
data_bundle, embed = load_data()
print(data_bundle)
model = TENER(tag_vocab=data_bundle.get_vocab('target'),
embed=embed,
num_layers=num_layers,
d_model=d_model,
n_head=n_heads,
feedforward_dim=dim_feedforward,
dropout=trans_dropout,
after_norm=after_norm,
attn_type=attn_type,
bi_embed=None,
fc_dropout=fc_dropout,
pos_embed=pos_embed,
scale=attn_type == 'naive')
if optim_type == 'sgd':
optimizer = optim.SGD(model.parameters(), lr=lr, momentum=0.9)
else:
optimizer = optim.Adam(model.parameters(), lr=lr, betas=(0.9, 0.99))
callbacks = []
clip_callback = GradientClipCallback(clip_type='value', clip_value=5)
evaluate_callback = EvaluateCallback(data_bundle.get_dataset('test'))
print("[Info] Zen Mode, Zen dataset loaded ...")
model = TENER(tag_vocab=data_bundle.get_vocab('target'),
embed=embed,
num_layers=num_layers,
d_model=d_model,
n_head=n_heads,
feedforward_dim=dim_feedforward,
dropout=args.trans_dropout,
after_norm=after_norm,
attn_type=attn_type,
bi_embed=bi_embed,
fc_dropout=fc_dropout,
pos_embed=pos_embed,
scale=attn_type == 'transformer',
use_knowledge=True,
feature2count=feature2count,
vocab_size=vocab_size,
feature_vocab_size=feature_vocab_size,
kv_attn_type=args.kv_attn_type,
memory_dropout=args.memory_dropout,
fusion_dropout=args.fusion_dropout,
fusion_type=args.fusion_type,
highway_layer=args.highway_layer,
key_embed_dropout=args.key_embed_dropout,
knowledge_type=args.knowledge_type,
use_zen=args.zen_model != "")
if args.optim_type == 'sgd':
optimizer = optim.SGD(model.parameters(), lr=lr, momentum=0.9)
requires_grad=True, lower=True, word_dropout=0, dropout=0.5,
only_norm_found_vector=normalize_embed)
if char_embed is not None:
embed = StackEmbedding([word_embed, char_embed], dropout=0, word_dropout=0.02)
else:
word_embed.word_drop = 0.02
embed = word_embed
data__ = data.get_vocab('words')
data.rename_field('words', 'chars')
return data, embed, data__, word_embed
data_bundle, embed, data__, word_embed = load_data()
model = TENER(tag_vocab=data_bundle.get_vocab('target'), embed=embed, num_layers=num_layers,
d_model=d_model, n_head=n_heads,
feedforward_dim=dim_feedforward, dropout=dropout,
after_norm=after_norm, attn_type=attn_type,
bi_embed=None,
fc_dropout=fc_dropout,
pos_embed=pos_embed,
scale=attn_type=='transformer')
#Thay model path trước khi chạy
model_path = './w2v_n6'
states = torch.load(model_path).state_dict()
model.load_state_dict(states)
evaluation = Tester(data_bundle.get_dataset('test'),
model,
metrics=SpanFPreRecMetric(tag_vocab=data_bundle.get_vocab('target'), only_gross=False),
batch_size=batch_size)
evaluation.test()