def __init__(self, opt):
self.opt = opt
self.tt = torch.cuda if opt.cuda else torch
checkpoint = torch.load(opt.model,
map_location=lambda storage, loc: storage)
model_opt = checkpoint['settings']
if 'use_ctx' not in model_opt.__dict__:
model_opt.use_ctx = False
self.model_opt = model_opt
model = Transformer(model_opt.src_vocab_size,
model_opt.tgt_vocab_size,
model_opt.max_token_seq_len,
proj_share_weight=model_opt.proj_share_weight,
embs_share_weight=model_opt.embs_share_weight,
d_k=model_opt.d_k,
d_v=model_opt.d_v,
d_model=model_opt.d_model,
d_word_vec=model_opt.d_word_vec,
d_inner_hid=model_opt.d_inner_hid,
n_layers=model_opt.n_layers,
n_head=model_opt.n_head,
dropout=model_opt.dropout,
use_ctx=model_opt.use_ctx)
prob_projection = nn.LogSoftmax()
model.load_state_dict(checkpoint['model'])
# New max_token_seq_len for position encoding
model = self.change_position_embedings(model, opt.max_token_seq_len,
model_opt.d_word_vec,
model_opt.use_ctx)
model_opt.max_token_seq_len = opt.max_token_seq_len
print('[Info] Trained model state loaded.')
if opt.cuda:
model.cuda()
prob_projection.cuda()
else:
model.cpu()
prob_projection.cpu()
model.prob_projection = prob_projection
self.model = model
self.model.eval()
def __init__(self, opt):
self.opt = opt
self.tt = torch.cuda if opt.cuda else torch
checkpoint = torch.load(opt.model)
model_opt = checkpoint['settings']
self.model_opt = model_opt
model = Transformer(
model_opt.src_vocab_size,
model_opt.tgt_vocab_size,
model_opt.max_token_seq_len,
proj_share_weight=model_opt.proj_share_weight,
embs_share_weight=model_opt.embs_share_weight,
d_k=model_opt.d_k,
d_v=model_opt.d_v,
d_model=model_opt.d_model,
d_word_vec=model_opt.d_word_vec,
d_inner_hid=model_opt.d_inner_hid,
n_layers=model_opt.n_layers,
n_head=model_opt.n_head,
dropout=model_opt.dropout)
prob_projection = nn.LogSoftmax()
model.load_state_dict(checkpoint['model'])
print('[Info] Trained model state loaded.')
if opt.cuda:
model.cuda()
prob_projection.cuda()
else:
model.cpu()
prob_projection.cpu()
model.prob_projection = prob_projection
self.model = model
self.model.eval()
