code
stringlengths
101
5.91M
_model def tf_efficientnetv2_m_in21k(pretrained=False, **kwargs): kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['pad_type'] = 'same' model = _gen_efficientnetv2_m('tf_efficientnetv2_m_in21k', pretrained=pretrained, **kwargs) return model
class RepresentationWarpEncoder(nn.Module): def __init__(self, *, ch, ch_mult, num_res_blocks, attn_resolutions, r_channels, dropout=0.0, in_channels, resolution, act='silu', num_heads=1, num_ch_per_head=None, embedding_type='fourier', fourier_scale=16.0, use_adaptive_group_norm=False, scale_by_sqrt2=False, num_classes=None, time_conditional=False): super().__init__() self.act = get_act(act) self.num_resolutions = len(ch_mult) self.num_res_blocks = num_res_blocks self.num_classes = num_classes self.embedding_type = embedding_type self._time_conditional = time_conditional if time_conditional: self.time_conditioning = nn.ModuleList() if (self.embedding_type.lower() == 'fourier'): self.time_conditioning.append(GaussianFourierProjection(embedding_size=ch, scale=fourier_scale)) embed_dim = (2 * ch) elif (self.embedding_type.lower() == 'positional'): self.time_conditioning.append(PositionalSinusoidalEmbedding(embedding_dim=ch)) embed_dim = ch else: raise ValueError(f"Invalid embedding_type. Must be one of: 'fourier', 'positional'. Was: '{self.embedding_type}'") self.time_conditioning.append(nn.Linear(embed_dim, (4 * ch))) self.time_conditioning.append(self.act) self.time_conditioning.append(nn.Linear((4 * ch), (4 * ch))) self.time_conditioning = nn.Sequential(*self.time_conditioning) self._class_conditional = (self.num_classes is not None) if self._class_conditional: self.label_emb = nn.Embedding(self.num_classes, (4 * ch)) _AttnBlock = functools.partial(AttentionBlock, num_heads=num_heads, num_ch_per_head=num_ch_per_head, scale_by_sqrt2=scale_by_sqrt2) _ResnetBlock = functools.partial(ResnetBlockBigGAN, act=self.act, emb_dim=(4 * ch), dropout=dropout, use_adaptive_group_norm=use_adaptive_group_norm, scale_by_sqrt2=scale_by_sqrt2) _Downsample = functools.partial(_ResnetBlock, down=True) self.conv_in = nn.Conv2d(in_channels, ch, 3, padding=1) self.down = nn.ModuleList() current_res = resolution in_ch = ch h_channels = [in_ch] for level in range(self.num_resolutions): stage = nn.Module() stage.main = nn.ModuleList() stage.uses_attn = (current_res in attn_resolutions) out_ch = (ch * ch_mult[level]) for _ in range(self.num_res_blocks): stage.main.append(_ResnetBlock(in_ch=in_ch, out_ch=out_ch)) if stage.uses_attn: stage.main.append(_AttnBlock(channels=out_ch)) h_channels.append(out_ch) in_ch = out_ch if (level != (self.num_resolutions - 1)): stage.downsample = _Downsample(in_ch=in_ch) current_res = (current_res // 2) h_channels.append(in_ch) self.down.append(stage) self.mid = nn.ModuleList([_ResnetBlock(in_ch=in_ch), _AttnBlock(channels=in_ch), _ResnetBlock(in_ch=in_ch)]) self.norm_out = nn.GroupNorm(num_groups=min((in_ch // 4), 32), num_channels=in_ch, eps=1e-06) self.conv_out = torch.nn.Conv2d(in_ch, (2 * r_channels), kernel_size=3, stride=1, padding=1) def time_conditional(self): return self._time_conditional def class_conditional(self): return self._class_conditional def forward(self, x, time_cond=None, y=None): emb = None if self.time_conditional: assert (time_cond is not None) assert (time_cond.shape == (x.shape[0],)) emb = self.time_conditioning(time_cond) if self.class_conditional: assert (y is not None), 'Missing class label for class-cond. model' assert (y.shape == (x.shape[0],)) if (emb is not None): emb = (emb + self.label_emb(y)) else: emb = self.label_emb(y) h = self.conv_in(x) for stage in self.down: for block in stage.main: if (stage.uses_attn and isinstance(block, AttentionBlock)): h = block(h) else: h = block(h, emb) if hasattr(stage, 'downsample'): h = stage.downsample(h, emb) for block in self.mid: if isinstance(block, AttentionBlock): h = block(h) else: h = block(h, emb) h = self.norm_out(h) h = self.act(h) h = self.conv_out(h) return h
def test_mildnonaxi_oortA_grid(): idf = dehnendf(beta=0.0) pot = [LogarithmicHaloPotential(normalize=1.0), EllipticalDiskPotential(twophio=0.001)] edf = evolveddiskdf(idf, pot=pot, to=(- 10.0)) (oa, grid, dgridR, dgridphi) = edf.oortA(0.9, phi=0.2, integrate_method='rk6_c', grid=True, derivRGrid=True, derivphiGrid=True, returnGrids=True, gridpoints=_GRIDPOINTS, derivGridpoints=_GRIDPOINTS) ioa = idf.oortA(0.9) assert (numpy.fabs((oa - ioa)) < 0.005), 'oortA of evolveddiskdf for axisymmetric potential is not equal to that of initial DF' oa = edf.oortA(0.9, phi=0.2, integrate_method='rk6_c', grid=grid, derivRGrid=dgridR, derivphiGrid=dgridphi, gridpoints=_GRIDPOINTS, derivGridpoints=_GRIDPOINTS) assert (numpy.fabs((oa - ioa)) < 0.005), 'oortA of evolveddiskdf for axisymmetric potential is not equal to that of initial DF when calculated with pre-computed grid' return None
class RandomAffine(): def __call__(self, image): image = numpy.asarray(image) random_hsv(image) image = random_affine(image) return Image.fromarray(image)
class OrderedSet(collections.MutableSet): def __init__(self, iterable=None): self.end = end = [] end += [None, end, end] self.map = {} if (iterable is not None): self |= iterable def __len__(self): return len(self.map) def __contains__(self, key): return (key in self.map) def add(self, key): if (key not in self.map): end = self.end curr = end[1] curr[2] = end[1] = self.map[key] = [key, curr, end] def discard(self, key): if (key in self.map): (key, prev, next) = self.map.pop(key) prev[2] = next next[1] = prev def __iter__(self): end = self.end curr = end[2] while (curr is not end): (yield curr[0]) curr = curr[2] def __reversed__(self): end = self.end curr = end[1] while (curr is not end): (yield curr[0]) curr = curr[1] def pop(self, last=True): if (not self): raise KeyError('set is empty') key = (self.end[1][0] if last else self.end[2][0]) self.discard(key) return key def __repr__(self): if (not self): return ('%s()' % (self.__class__.__name__,)) return ('%s(%r)' % (self.__class__.__name__, list(self))) def __eq__(self, other): if isinstance(other, OrderedSet): return ((len(self) == len(other)) and (list(self) == list(other))) return (set(self) == set(other))
def load_model_(model_path): check = 'model_/checkpoint_epoch_168_val_loss_0._dice_0..pth' network = unet3d.UNet3D(in_channels=4, out_channels=4, final_sigmoid=False, f_maps=32, layer_order='crg', num_levels=4, num_groups=4, conv_padding=1) return _load(network, model_path, check)
def test_reset_scorer() -> None: with pytest.raises(ValueError, match='useless is not a valid scorer '): get_scorer('useless') register_scorer('useless', make_scorer(_return_1)) get_scorer('useless') reset_scorer_register() with pytest.raises(ValueError, match='useless is not a valid scorer '): get_scorer('useless')
class TestKLDiv(unittest.TestCase): def setUp(self) -> None: self.shape = (10, 5, 224, 224) self.logit = torch.randn(*self.shape, requires_grad=True) self.pred = F.softmax(self.logit, 1) self.target = torch.randint(low=0, high=self.shape[1], size=[self.shape[i] for i in range(self.shape.__len__()) if (i != 1)]) self.target_oh = class2one_hot(self.target, C=self.shape[1]).float() def _test_kl_equivalent(self, reduction='mean'): kl_criterion = nn.KLDivLoss(reduction=reduction) kl_loss = kl_criterion(self.pred.log(), target=self.target_oh) _kl_loss = loss.KL_div(reduction=reduction)(self.pred, self.target_oh) assert torch.isclose(kl_loss, ((_kl_loss / self.shape[1]) if (reduction == 'mean') else _kl_loss)) def test_kl_equivalent(self): for reduction in ('sum', 'mean'): self._test_kl_equivalent(reduction=reduction) def test_entropy(self): random_entropy = loss.Entropy()(self.pred) with self.assertRaises(AssertionError): loss.Entropy()(self.logit) max_entropy = loss.Entropy()(torch.zeros_like(self.pred).fill_((1 / self.shape[1]))) assert (random_entropy <= max_entropy) zero_entropy = loss.Entropy()(torch.Tensor([[1, 0], [0, 1]])) assert (zero_entropy == 0)
def neuralchat_client_execute(): com = neuralchat_client_commands idx = 0 for _argv in (['neuralchat_client'] + sys.argv[1:]): if (_argv not in com): break idx += 1 com = com[_argv] if (not callable(com['_command'])): i = com['_command'].rindex('.') (module, cls) = (com['_command'][:i], com['_command'][(i + 1):]) exec('from {} import {}'.format(module, cls)) com['_command'] = locals()[cls] status = (0 if com['_command']().execute(sys.argv[idx:]) else 1) return status
def mdetr_efficientnetB3(pretrained=False, return_postprocessor=False): model = _make_detr('timm_tf_efficientnet_b3_ns') if pretrained: checkpoint = torch.hub.load_state_dict_from_url(url=' map_location='cpu', check_hash=True) model.load_state_dict(checkpoint['model']) if return_postprocessor: return (model, PostProcess()) return model
def ODETimeSplitter(num_split, K): eta = (K / num_split) return [(i * eta) for i in range(1, num_split)]
_model def semnasnet_075(pretrained=False, **kwargs): model = _gen_mnasnet_a1('semnasnet_075', 0.75, pretrained=pretrained, **kwargs) return model
class PreResUnit(nn.Module): def __init__(self, in_channels, out_channels, stride, bottleneck, conv1_stride): super(PreResUnit, self).__init__() self.resize_identity = ((in_channels != out_channels) or (stride != 1)) if bottleneck: self.body = PreResBottleneck(in_channels=in_channels, out_channels=out_channels, stride=stride, conv1_stride=conv1_stride) else: self.body = PreResBlock(in_channels=in_channels, out_channels=out_channels, stride=stride) if self.resize_identity: self.identity_conv = conv1x1(in_channels=in_channels, out_channels=out_channels, stride=stride) def forward(self, x): identity = x (x, x_pre_activ) = self.body(x) if self.resize_identity: identity = self.identity_conv(x_pre_activ) x = (x + identity) return x
def _replace_end(string, end, new_end): if (string == end): return new_end if string.endswith(('.' + end)): return ((string[:(- len(('.' + end)))] + '.') + new_end) return string
_cache(maxsize=100) def get_tiny_tokenizer_from_checkpoint(checkpoint): tokenizer = AutoTokenizer.from_pretrained(checkpoint) if (tokenizer.vocab_size < 300): return tokenizer logger.info('Training new from iterator ...') vocabulary = ((string.ascii_letters + string.digits) + ' ') tokenizer = tokenizer.train_new_from_iterator(vocabulary, vocab_size=len(vocabulary), show_progress=False) logger.info('Trained.') return tokenizer
class BirdsRefTask(RefTask): def __init__(self): super(BirdsRefTask, self).__init__(N_FEATURES) with open('data/birds/labels.p') as feature_f: features = pickle.load(feature_f) self.random = np.random.RandomState(0) (n_raw_features,) = features.values()[0].shape projector = self.random.randn(N_FEATURES, n_raw_features) proj_features = {} m1 = np.zeros(N_FEATURES) m2 = np.zeros(N_FEATURES) for (k, v) in features.items(): proj = v m1 += proj m2 += (proj ** 2) proj_features[k] = proj m1 /= len(features) m2 /= len(features) std = np.sqrt((m2 - (m1 ** 2))) self.features = proj_features self.keys = sorted(self.features.keys()) self.folds = {} for fold in ['train', 'val', 'test']: with open(('data/birds/%s.txt' % fold)) as fold_f: self.folds[fold] = [line.strip() for line in fold_f] anns = {} with open('data/birds/cub_0917_5cap.tsv') as ann_f: reader = csv.reader(ann_f, delimiter='\t') header = reader.next() i_url = header.index('Input.image_url') i_desc = header.index('Answer.Description') for line in reader: url = line[i_url] desc = line[i_desc] key = '/'.join(url.split('/')[(- 2):]) desc = desc.lower().replace('.', '').replace(',', '') anns[key] = tuple(desc.split()) counts = defaultdict((lambda : 0)) for desc in anns.values(): for i in range((len(desc) - 1)): bigram = desc[i:(i + 2)] counts[bigram] += 1 freq_terms = sorted(counts.items(), key=(lambda p: (- p[1]))) freq_terms = [f for f in freq_terms if (not any(((w in STOP) for w in f[0])))] freq_terms = [f[0] for f in freq_terms] freq_terms = freq_terms[:50] self.vocab = {'_': 0, 'UNK': 1} self.reverse_vocab = {0: '_', 1: 'UNK'} self.lexicon = [[0]] for term in freq_terms: for word in term: if (word in self.vocab): continue index = len(self.vocab) self.vocab[word] = index self.reverse_vocab[index] = word self.lexicon.append([self.vocab[w] for w in term]) discarded = [] self.reps = {} for (k, desc) in anns.items(): rep = np.zeros(len(self.lexicon)) out = [] for (i_l, l) in enumerate(self.lexicon): if all(((self.reverse_vocab[w] in desc) for w in l)): rep[i_l] = 1 out += l if (len(out) == 0): discarded.append(k) continue assert rep.any() rep /= np.sum(rep) self.reps[k] = rep for k in discarded: del anns[k] for fold in self.folds.values(): if (k in fold): fold.remove(k) self.empty_desc = np.zeros(len(self.lexicon)) self.empty_desc[0] = 1 def reset_test(self): self.random = np.random.RandomState(0) def get_pair(self, fold): fold_keys = self.folds[fold] (i1, i2) = self.random.randint(len(fold_keys), size=2) (k1, k2) = (fold_keys[i1], fold_keys[i2]) (f1, f2) = (self.features[k1], self.features[k2]) rep = self.reps[k1] return (f1, f2, rep, k1, k2) def visualize(self, state, agent): url_template = ' url1 = (url_template % state.left_data) url2 = (url_template % state.right_data) if ((agent == 0) and (state.target == 1)): (url1, url2) = (url2, url1) html_template = "<img src='%s'>" return ((html_template % url1) + (html_template % url2)) def turk_visualize(self, state, agent, loc): left = state.left_data right = state.right_data if ((agent == 0) and (state.target == 1)): (left, right) = (right, left) return (left, right) def pp(self, indices): return ' '.join([self.reverse_vocab[i] for i in indices])
def load_dict_classifiers(args, classifiers_name=DEFAULT_LIST_CLASSIFIERS): dict_classifier = load_list_classifiers(args, classifiers_name=classifiers_name) (l_train, l_test) = split_classif_list(args, dict_classifier) print(f''' Training on {l_train.keys()} ''') print(f''' Testing on {l_test.keys()} ''') if args.ensemble_adv_trained: l_test = load_ens_adv_model(args) return (l_train, l_test)
def training_user_task(model, sess): best_loss = 0 saver = tf.train.Saver() ts_u = data.generate_user_dict_train(setting.oracle_training_file_user_task) vs_u = data.generate_user_dict_valid(setting.oracle_valid_file_user_task) train_batches = data.generate_meta_train_user_set(ts_u) valid_batches = data.generate_meta_valid_user_set(vs_u) num_batch = (int(train_batches[3]) // setting.batch_size_user) batch_index = range(num_batch) valid_num_batch = (int(valid_batches[3]) // setting.batch_size_user) valid_batch_index = range(valid_num_batch) for epoch_count in range(setting.user_epoch): train_begin = time() training_batch_user_task(batch_index, model, sess, train_batches, True) train_time = (time() - train_begin) if ((epoch_count % setting.verbose) == 0): loss_begin = time() train_loss = training_loss_user_task(batch_index, model, sess, train_batches, True) loss_time = (time() - loss_begin) eval_begin = time() cosine = evaluate_user_task(valid_batch_index, model, sess, valid_batches, True) eval_time = (time() - eval_begin) print(('epoch %d, train time is %.4f, loss time is %.4f, eval_time is %.4f, train_loss is %.4f, test cosine value is %.4f' % (epoch_count, train_time, loss_time, eval_time, train_loss, cosine))) if (cosine < best_loss): best_loss = cosine saver.save(sess, setting.checkpoint_path_user_task, global_step=epoch_count) ts_u = data.generate_user_dict_train(setting.oracle_training_file_user_task) vs_u = data.generate_user_dict_valid(setting.oracle_valid_file_user_task) train_batches = data.generate_meta_train_user_set(ts_u) valid_batches = data.generate_meta_valid_user_set(vs_u)
class TestTransitDepthCalculator(unittest.TestCase): def get_frac_dev(self, logZ, CO_ratio, custom_abundances): Rp = .0 Mp = 2e+27 Rs = .0 T = 1200 depth_calculator = TransitDepthCalculator() (wavelengths, transit_depths) = depth_calculator.compute_depths(Rs, Mp, Rp, T, logZ=logZ, CO_ratio=CO_ratio, custom_abundances=custom_abundances, cloudtop_pressure=10000.0) (ref_wavelengths, ref_depths) = np.loadtxt('tests/testing_data/hot_jupiter_spectra.dat', unpack=True, skiprows=2) ref_depths /= 100 frac_dev = (np.abs((ref_depths - transit_depths)) / ref_depths) return frac_dev def test_custom_file(self): custom_abundances = AbundanceGetter.from_file('tests/testing_data/abund_1Xsolar_cond.dat') for key in ['HCl', 'HF', 'MgH', 'SH', 'SiH', 'C2H2', 'C2H4', 'C2H6', 'H2CO', 'OCS']: del custom_abundances[key] frac_dev = self.get_frac_dev(None, None, custom_abundances) self.assertLess(np.percentile(frac_dev, 95), 0.03) self.assertLess(np.max(frac_dev), 0.07) def test_ggchem(self): frac_dev = self.get_frac_dev(0, 0.53, None) self.assertLess(np.percentile(frac_dev, 95), 0.03) self.assertLess(np.max(frac_dev), 0.1) def test_unbound_atmosphere(self): Rp = 6378000.0 Mp = 5.97e+20 Rs = .0 T = 300 depth_calculator = TransitDepthCalculator() with self.assertRaises(AtmosphereError): (wavelengths, transit_depths) = depth_calculator.compute_depths(Rs, Mp, Rp, T, logZ=0.2, CO_ratio=1.1, T_star=6100) def test_bin_wavelengths(self): Rp = .0 Mp = 7.49e+26 Rs = .0 T = 1200 depth_calculator = TransitDepthCalculator() bins = np.array([[0.4, 0.6], [1, 1.1], [1.2, 1.4], [3.2, 4], [5, 6]]) bins *= 1e-06 depth_calculator.change_wavelength_bins(bins) (wavelengths, transit_depths) = depth_calculator.compute_depths(Rs, Mp, Rp, T, logZ=0.2, CO_ratio=1.1, T_star=6100) self.assertEqual(len(wavelengths), len(bins)) self.assertEqual(len(transit_depths), len(bins)) (wavelengths, transit_depths) = depth_calculator.compute_depths(Rs, Mp, Rp, T, logZ=0.2, CO_ratio=1.1, T_star=12000) self.assertEqual(len(wavelengths), len(bins)) self.assertEqual(len(transit_depths), len(bins)) def test_power_law_haze(self): Rs = R_sun Mp = M_jup Rp = R_jup T = 1200 abundances = AbundanceGetter().get(0, 0.53) for key in abundances: abundances[key] *= 0 abundances['H2'] += 1 depth_calculator = TransitDepthCalculator() (wavelengths, transit_depths, info_dict) = depth_calculator.compute_depths(Rs, Mp, Rp, T, logZ=None, CO_ratio=None, cloudtop_pressure=np.inf, custom_abundances=abundances, add_gas_absorption=False, add_collisional_absorption=False, full_output=True) g = ((G * Mp) / (Rp ** 2)) H = ((k_B * T) / ((2 * AMU) * g)) gamma = 0.57721 polarizability = 8.059e-31 sigma = ((((128.0 * (np.pi ** 5)) / 3) * (polarizability ** 2)) / (depth_calculator.atm.lambda_grid ** 4)) kappa = (sigma / (2 * AMU)) P_surface = .0 R_surface = info_dict['radii'][(- 1)] tau_surface = (((P_surface / g) * np.sqrt((((2 * np.pi) * R_surface) / H))) * kappa) analytic_R = (R_surface + (H * ((gamma + np.log(tau_surface)) + scipy.special.expn(1, tau_surface)))) analytic_depths = ((analytic_R ** 2) / (Rs ** 2)) ratios = (analytic_depths / transit_depths) relative_diffs = np.abs((ratios - 1)) self.assertTrue(np.all((relative_diffs < 0.001))) def test_k_coeffs_unbinned(self): xsec_calc = TransitDepthCalculator(method='xsec') ktab_calc = TransitDepthCalculator(method='ktables') (xsec_wavelengths, xsec_depths) = xsec_calc.compute_depths(R_sun, M_jup, R_jup, 1000) N = 10 smoothed_xsec_wavelengths = uniform_filter(xsec_wavelengths, N)[::N] smoothed_xsec_depths = uniform_filter(xsec_depths, N)[::N] (ktab_wavelengths, ktab_depths) = ktab_calc.compute_depths(R_sun, M_jup, R_jup, 1000) diffs = np.abs((ktab_depths - smoothed_xsec_depths[:(- 1)])) self.assertTrue((np.median(diffs) < 2e-05)) self.assertTrue((np.percentile(diffs, 95) < 5e-05)) self.assertTrue((np.max(diffs) < 0.00015)) def test_k_coeffs_binned(self): wavelengths = np.exp(np.arange(np.log(3.1e-07), np.log(2.9e-05), (1.0 / 20))) wavelength_bins = np.array([wavelengths[0:(- 1)], wavelengths[1:]]).T xsec_calc = TransitDepthCalculator(method='xsec') xsec_calc.change_wavelength_bins(wavelength_bins) ktab_calc = TransitDepthCalculator(method='ktables') ktab_calc.change_wavelength_bins(wavelength_bins) (wavelengths, xsec_depths) = xsec_calc.compute_depths(R_sun, M_jup, R_jup, 300, logZ=1, CO_ratio=1.5) (wavelengths, ktab_depths) = ktab_calc.compute_depths(R_sun, M_jup, R_jup, 300, logZ=1, CO_ratio=1.5) diffs = np.abs((ktab_depths - xsec_depths)) self.assertTrue((np.median(diffs) < 1e-05)) self.assertTrue((np.percentile(diffs, 95) < 2e-05)) self.assertTrue((np.max(diffs) < 3e-05)) def test_bounds_checking(self): Rp = .0 Mp = 7.49e+26 Rs = .0 T = 1200 logZ = 0 CO_ratio = 1.1 calculator = TransitDepthCalculator() with self.assertRaises(AtmosphereError): calculator.compute_depths(Rs, Mp, Rp, 199, logZ=logZ, CO_ratio=CO_ratio) with self.assertRaises(AtmosphereError): calculator.compute_depths(Rs, Mp, Rp, 3001, logZ=logZ, CO_ratio=CO_ratio) with self.assertRaises(ValueError): calculator.compute_depths(Rs, Mp, Rp, T, logZ=(- 1.1), CO_ratio=CO_ratio) with self.assertRaises(ValueError): calculator.compute_depths(Rs, Mp, Rp, T, logZ=3.1, CO_ratio=CO_ratio) with self.assertRaises(ValueError): calculator.compute_depths(Rs, Mp, Rp, T, logZ=logZ, CO_ratio=0.01) with self.assertRaises(ValueError): calculator.compute_depths(Rs, Mp, Rp, T, logZ=logZ, CO_ratio=11) with self.assertRaises(ValueError): calculator.compute_depths(Rs, Mp, Rp, T, logZ=logZ, CO_ratio=CO_ratio, cloudtop_pressure=0.0001) with self.assertRaises(ValueError): calculator.compute_depths(Rs, Mp, Rp, T, logZ=logZ, CO_ratio=CO_ratio, cloudtop_pressure=.0) calculator.compute_depths(Rs, Mp, Rp, T, logZ=logZ, CO_ratio=CO_ratio, cloudtop_pressure=np.inf)
def preprocess_data(args): if (args['task_name'] == 'paws_labeled_final_wiki'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/paws', 'labeled_final', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'paws_labeled_swap_wiki'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/paws', 'labeled_swap', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'paws_unlabeled_final_wiki'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/paws', 'unlabeled_final', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'paws_qqp'): data_0 = pd.read_csv('/yanglinyi/new_codes_cluster/paws/paws_qqp/output/train.tsv', sep='\t') data_1 = pd.read_csv('/yanglinyi/new_codes_cluster/paws/paws_qqp/output/dev_and_test.tsv', sep='\t') data = pd.concat([data_0, data_1], ignore_index=True) data = da.Dataset.from_pandas(data) elif (args['task_name'] == 'qqp'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/glue', 'qqp', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'mrpc'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/glue', 'mrpc', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'rte'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/glue', 'rte', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'mnli_mismatched'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/glue', 'mnli', split='validation_mismatched', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'sick_num5'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/sick', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'sick_num3'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/sick', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'hans'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/hans', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'qnli_validation'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/glue', 'qnli', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') data = data.rename_column('question', 'sentence1') data = data.rename_column('sentence', 'sentence2') elif (args['task_name'] == 'cola_validation'): data = load_dataset('/yanglinyi/new_codes_cluster/datasets/glue', 'cola', cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') elif (args['task_name'] == 'cola_ood'): from datasets import load_from_disk data = load_from_disk('/yanglinyi/new_codes_cluster/cola_data/cola_ood') elif (args['task_name'] == 'qnli_ood'): from datasets import load_from_disk data = load_from_disk('/yanglinyi/new_codes_cluster/qnli_data/qnli_ood') elif (args['task_name'] == 'scitail'): data = pd.read_csv('/yanglinyi/new_codes_cluster/datasets/scitail/merged_scitail.tsv', sep='\t') data = da.Dataset.from_pandas(data) elif (args['task_name'] == 'flipkart'): data = pd.read_csv('/yanglinyi/new_codes_cluster/datasets/flipkart/flipkart_updated.tsv', sep='\t', error_bad_lines=False) data = da.Dataset.from_pandas(data) elif (args['task_name'] == 'twitter'): data_train = pd.read_csv('/yanglinyi/new_codes_cluster/datasets/SemEval-PIT2015-github/train.tsv', sep='\t', error_bad_lines=False) data_dev = pd.read_csv('/yanglinyi/new_codes_cluster/datasets/SemEval-PIT2015-github/dev.tsv', sep='\t', error_bad_lines=False) data_test = pd.read_csv('/yanglinyi/new_codes_cluster/datasets/SemEval-PIT2015-github/test.tsv', sep='\t', error_bad_lines=False) data = pd.concat([data_train, data_dev, data_test], ignore_index=True) data = da.Dataset.from_pandas(data) else: data = load_dataset(('/yanglinyi/new_codes_cluster/datasets/' + args['task_name']), cache_dir='/yanglinyi/new_codes_cluster/huggingface/datasets') if (args['task_name'] == 'sick_num5'): data = data.map((lambda examples: {'relatedness_score': round((examples['relatedness_score'] - 1.0))})) data = data.rename_column('sentence_A', 'sentence1') data = data.rename_column('sentence_B', 'sentence2') data = data.remove_columns(['id', 'label', 'entailment_AB', 'entailment_BA', 'sentence_A_original', 'sentence_B_original', 'sentence_A_dataset', 'sentence_B_dataset']) data = data.rename_column('relatedness_score', 'label') if (args['task_name'] == 'sick_num3'): data = data.rename_column('sentence_A', 'sentence1') data = data.rename_column('sentence_B', 'sentence2') data = data.remove_columns(['id', 'relatedness_score', 'entailment_AB', 'entailment_BA', 'sentence_A_original', 'sentence_B_original', 'sentence_A_dataset', 'sentence_B_dataset']) if (args['task_name'] == 'imdb'): data = data.rename_column('text', 'sentence') elif (args['task_name'] == 'yelp_polarity'): data = data.rename_column('text', 'sentence') elif (args['task_name'] == 'amazon_polarity'): data = data.rename_column('content', 'sentence') data = data.remove_columns('title') elif (args['task_name'] == 'qqp'): if (args['ID_name'] == 'rte'): data = data.map((lambda examples: {'label': ((- examples['label']) + 1.0)})) data = data.rename_column('question1', 'sentence1') data = data.rename_column('question2', 'sentence2') elif (args['task_name'] == 'rte'): if (args['ID_name'] != 'rte'): data = data.map((lambda examples: {'label': ((- examples['label']) + 1.0)})) elif (args['task_name'] == 'mnli_mismatched'): data = data.rename_column('hypothesis', 'sentence2') data = data.rename_column('premise', 'sentence1') elif (args['task_name'] == 'snli'): data = data.rename_column('hypothesis', 'sentence2') data = data.rename_column('premise', 'sentence1') elif (args['task_name'] == 'mrpc'): if (args['ID_name'] == 'rte'): data = data.map((lambda examples: {'label': ((- examples['label']) + 1.0)})) elif (args['task_name'] == 'hans'): if (args['ID_name'] != 'rte'): data = data.map((lambda examples: {'label': ((- examples['label']) + 1.0)})) data = data.rename_column('premise', 'sentence1') data = data.rename_column('hypothesis', 'sentence2') data = data.remove_columns(['parse_premise', 'parse_hypothesis', 'binary_parse_premise', 'binary_parse_hypothesis', 'heuristic', 'subcase', 'template']) elif ('paws' in args['task_name']): if (args['ID_name'] == 'rte'): data = data.map((lambda examples: {'label': ((- examples['label']) + 1.0)})) elif (args['task_name'] == 'qnli_ood'): data = data.rename_column('question', 'sentence1') data = data.rename_column('sentence', 'sentence2') def convert_qnli_dataset(example): if (example['label'] == 'entailment'): return 0 elif (example['label'] == 'not_entailment'): return 1 else: raise ValueError('label not legal') data = data.map((lambda examples: {'label': convert_qnli_dataset(examples)})) elif (args['task_name'] == 'scitail'): def convert_scitail_dataset(example): if (example['label'] == 'entails'): return 0 elif (example['label'] == 'neutral'): return 1 else: raise ValueError('label not legal') data = data.map((lambda examples: {'label': convert_scitail_dataset(examples)})) data = data.rename_column('premise', 'sentence1') data = data.rename_column('hypothesis', 'sentence2') elif (args['task_name'] == 'flipkart'): def convert_flipkart_dataset(example): if (example['label'] == 'negative'): return 0 elif (example['label'] == 'neutral'): return 0 elif (example['label'] == 'positive'): return 1 else: raise ValueError('label not legal') data = data.rename_column('Summary', 'sentence') data = data.rename_column('Sentiment', 'label') data = data.map((lambda examples: {'label': convert_flipkart_dataset(examples)})) elif (args['task_name'] == 'twitter'): data = data.rename_column('Sent_1', 'sentence1') data = data.rename_column('Sent_2', 'sentence2') data = data.rename_column('Label', 'label') if ((args['task_name'] == 'imdb') or (args['task_name'] == 'yelp_polarity') or (args['task_name'] == 'amazon_polarity')): data = concatenate_datasets([data['train'], data['test']]) elif (args['task_name'] == 'paws_labeled_final_wiki'): data = concatenate_datasets([data['train'], data['test'], data['validation']]) elif (args['task_name'] == 'paws_labeled_swap_wiki'): data = data['train'] elif (args['task_name'] == 'paws_unlabeled_final_wiki'): data = concatenate_datasets([data['train'], data['validation']]) elif (args['task_name'] == 'paws_qqp'): data = data elif (args['task_name'] == 'sick_num5'): data = concatenate_datasets([data['train'], data['test'], data['validation']]) elif (args['task_name'] == 'sick_num3'): data = concatenate_datasets([data['train'], data['test'], data['validation']]) elif (args['task_name'] == 'qqp'): data = concatenate_datasets([data['train'], data['validation']]) elif (args['task_name'] == 'rte'): data = concatenate_datasets([data['train'], data['validation']]) elif (args['task_name'] == 'mrpc'): data = concatenate_datasets([data['train'], data['validation']]) elif (args['task_name'] == 'mnli_mismatched'): data = data elif (args['task_name'] == 'snli'): data = concatenate_datasets([data['train'], data['test'], data['validation']]) elif (args['task_name'] == 'hans'): data = concatenate_datasets([data['train'], data['validation']]) elif (args['task_name'] == 'qnli_validation'): data = data['validation'] elif (args['task_name'] == 'cola_validation'): data = data['validation'] elif (args['task_name'] == 'cola_ood'): data = data elif (args['task_name'] == 'qnli_ood'): data = data return data
_group.command('add') ('env') ('endpoint') def add_environment(env, endpoint): click.echo(f'Adding environment:') click.echo(f' ENVIRONMENT: {env}') click.echo(f' ENDPOINT: {endpoint}') add_env(env, endpoint)
def gen_layer(C_in, C_out, model_index): swap = False if ((model_index >= 0) and (model_index <= 251)): prim_index = (model_index // 63) model_index = (model_index % 63) conv_index = (((model_index // len(genotypes.ACTIVATION)) // len(genotypes.KERNEL_SIZE)) % len(genotypes.UPSAMPLE_PRIMITIVE)) kernel_index = ((model_index // len(genotypes.ACTIVATION)) % len(genotypes.KERNEL_SIZE)) act_index = (model_index % len(genotypes.ACTIVATION)) if ((model_index >= 60) and (model_index <= 62)): conv_index = 5 ' DepthToSpace - Second ' if ((model_index >= 252) and (model_index <= 311)): swap = True prim_index = ((model_index - 63) // 63) model_index = (model_index % 63) conv_index = (((model_index // len(genotypes.ACTIVATION)) // len(genotypes.KERNEL_SIZE)) % len(genotypes.UPSAMPLE_PRIMITIVE)) kernel_index = ((model_index // len(genotypes.ACTIVATION)) % len(genotypes.KERNEL_SIZE)) act_index = (model_index % len(genotypes.ACTIVATION)) ' Transposed Convolution ' if ((model_index >= 312) and (model_index <= 323)): prim_index = 4 conv_index = 5 kernel_index = ((model_index // len(genotypes.ACTIVATION)) % len(genotypes.KERNEL_SIZE)) act_index = (model_index % len(genotypes.ACTIVATION)) prim_op = genotypes.UPSAMPLE_PRIMITIVE[prim_index] conv_op = genotypes.UPSAMPLE_CONV[conv_index] kernel_size = genotypes.KERNEL_SIZE[kernel_index] act_op = genotypes.ACTIVATION[act_index] if (prim_op == 'pixel_shuffle'): if (not swap): prim_op_layer = operations.UPSAMPLE_PRIMITIVE_OPS[prim_op](C_in=C_in, C_out=C_out, kernel_size=kernel_size, act_op=act_op) conv_op_layer = operations.UPSAMPLE_CONV_OPS[conv_op](C_in=int((C_in / 4)), C_out=C_out, kernel_size=kernel_size, act_op=act_op) return nn.Sequential(prim_op_layer, conv_op_layer) else: conv_op_layer = operations.UPSAMPLE_CONV_OPS[conv_op](C_in=C_in, C_out=int((C_out * 4)), kernel_size=kernel_size, act_op=act_op) prim_op_layer = operations.UPSAMPLE_PRIMITIVE_OPS[prim_op](C_in=C_in, C_out=C_out, kernel_size=kernel_size, act_op=act_op) return nn.Sequential(conv_op_layer, prim_op_layer) else: prim_op_layer = operations.UPSAMPLE_PRIMITIVE_OPS[prim_op](C_in=C_in, C_out=C_out, kernel_size=kernel_size, act_op=act_op) conv_op_layer = operations.UPSAMPLE_CONV_OPS[conv_op](C_in=C_in, C_out=C_out, kernel_size=kernel_size, act_op=act_op) return nn.Sequential(prim_op_layer, conv_op_layer)
class EncoderDisp(nn.Module): def __init__(self, bott_channels, out_channels, bottleneck): super(EncoderDisp, self).__init__() self.bottleneck = bottleneck self.disp = nn.Sequential(nn.Conv2d(bott_channels, out_channels, 3, 1, 1, padding_mode='reflect'), nn.Sigmoid()) def forward(self, inputs): features = self.bottleneck(inputs) out = self.disp(features) return out
def clipped_audio(x, num_frames=c.NUM_FRAMES): if (x.shape[0] > num_frames): bias = np.random.randint(0, (x.shape[0] - num_frames)) clipped_x = x[bias:(num_frames + bias)] else: clipped_x = x return clipped_x
class ElasticTransformPseudo2D(DualTransform): def __init__(self, alpha=1000, sigma=50, alpha_affine=1, approximate=False, always_apply=False, p=0.5): super().__init__(always_apply, p) self.alpha = alpha self.sigma = sigma self.alpha_affine = alpha_affine self.approximate = approximate def apply(self, img, random_state=None): return F.elastic_transform_pseudo2D(img, self.alpha, self.sigma, self.alpha_affine, interpolation=cv2.INTER_LINEAR, border_mode=cv2.BORDER_REFLECT_101, value=None, random_state=random_state, approximate=False) def apply_to_mask(self, img, random_state=None): return F.elastic_transform_pseudo2D(img, self.alpha, self.sigma, self.alpha_affine, interpolation=cv2.INTER_NEAREST, border_mode=cv2.BORDER_REFLECT_101, value=None, random_state=random_state, approximate=False) def get_params(self, **data): return {'random_state': random.randint(0, 10000)}
class ResNet(object): __shared__ = ['norm_type', 'freeze_norm', 'weight_prefix_name'] def __init__(self, depth=50, freeze_at=2, norm_type='affine_channel', freeze_norm=True, norm_decay=0.0, variant='b', feature_maps=[2, 3, 4, 5], dcn_v2_stages=[], weight_prefix_name='', nonlocal_stages=[], gcb_stages=[], gcb_params=dict(), sac_stages=[], sac_params=None, rfp_stages=[], lr_mult_list=[1.0, 1.0, 1.0, 1.0], eca_k_size=[]): super(ResNet, self).__init__() if isinstance(feature_maps, Integral): feature_maps = [feature_maps] assert (depth in [18, 34, 50, 101, 152, 200]), 'depth {} not in [18, 34, 50, 101, 152, 200]' assert (variant in ['a', 'b', 'c', 'd']), 'invalid ResNet variant' assert (0 <= freeze_at <= 4), 'freeze_at should be 0, 1, 2, 3 or 4' assert (len(feature_maps) > 0), 'need one or more feature maps' assert (norm_type in ['bn', 'sync_bn', 'affine_channel']) assert (not ((len(nonlocal_stages) > 0) and (depth < 50))), 'non-local is not supported for resnet18 or resnet34' assert (len(lr_mult_list) == 4), 'lr_mult_list length must be 4 but got {}'.format(len(lr_mult_list)) self.depth = depth self.freeze_at = freeze_at self.norm_type = norm_type self.norm_decay = norm_decay self.freeze_norm = freeze_norm self.variant = variant self._model_type = 'ResNet' self.feature_maps = feature_maps self.dcn_v2_stages = dcn_v2_stages self.depth_cfg = {18: ([2, 2, 2, 2], self.basicblock), 34: ([3, 4, 6, 3], self.basicblock), 50: ([3, 4, 6, 3], self.bottleneck), 101: ([3, 4, 23, 3], self.bottleneck), 152: ([3, 8, 36, 3], self.bottleneck), 200: ([3, 12, 48, 3], self.bottleneck)} self.stage_filters = [64, 128, 256, 512] self._c1_out_chan_num = 64 self.na = NameAdapter(self) self.prefix_name = weight_prefix_name self.nonlocal_stages = nonlocal_stages self.nonlocal_mod_cfg = {50: 2, 101: 5, 152: 8, 200: 12} self.gcb_stages = gcb_stages self.gcb_params = gcb_params self.sac_stages = sac_stages self.sac_params = sac_params self.rfp_stages = rfp_stages self.with_sac = False if ((self.sac_params is not None) and (len(self.sac_stages) > 0)): self.with_sac = True self.lr_mult_list = lr_mult_list self.with_eca = (len(eca_k_size) == 4) self.eca_k_size = eca_k_size self.stage_num = (- 1) def _conv_offset(self, input, filter_size, stride, padding, act=None, name=None): out_channel = ((filter_size * filter_size) * 3) out = fluid.layers.conv2d(input, num_filters=out_channel, filter_size=filter_size, stride=stride, padding=padding, param_attr=ParamAttr(initializer=Constant(0.0), name=(name + '.w_0')), bias_attr=ParamAttr(initializer=Constant(0.0), name=(name + '.b_0')), act=act, name=name) return out def _conv_norm(self, input, num_filters, filter_size, stride=1, groups=1, act=None, name=None, dcn_v2=False, with_sac=False): _name = ((self.prefix_name + name) if (self.prefix_name != '') else name) lr_mult = 1.0 mult_idx = max((self.stage_num - 2), 0) mult_idx = min((self.stage_num - 2), 3) lr_mult = self.lr_mult_list[mult_idx] if with_sac: conv = SAConv2d(input, num_filters=num_filters, filter_size=filter_size, stride=stride, padding=((filter_size - 1) // 2), groups=groups, act=None, bias_attr=False, name=_name, lr_mult=lr_mult, **self.sac_params) elif (not dcn_v2): conv = fluid.layers.conv2d(input=input, num_filters=num_filters, filter_size=filter_size, stride=stride, padding=((filter_size - 1) // 2), groups=groups, act=None, param_attr=ParamAttr(name=(_name + '_weights'), learning_rate=lr_mult), bias_attr=False, name=(_name + '.conv2d.output.1')) else: offset_mask = self._conv_offset(input=input, filter_size=filter_size, stride=stride, padding=((filter_size - 1) // 2), act=None, name=(_name + '_conv_offset')) offset_channel = ((filter_size ** 2) * 2) mask_channel = (filter_size ** 2) (offset, mask) = fluid.layers.split(input=offset_mask, num_or_sections=[offset_channel, mask_channel], dim=1) mask = fluid.layers.sigmoid(mask) conv = fluid.layers.deformable_conv(input=input, offset=offset, mask=mask, num_filters=num_filters, filter_size=filter_size, stride=stride, padding=((filter_size - 1) // 2), groups=groups, deformable_groups=1, im2col_step=1, param_attr=ParamAttr(name=(_name + '_weights'), learning_rate=lr_mult), bias_attr=False, name=(_name + '.conv2d.output.1')) bn_name = self.na.fix_conv_norm_name(name) bn_name = ((self.prefix_name + bn_name) if (self.prefix_name != '') else bn_name) norm_lr = (0.0 if self.freeze_norm else lr_mult) norm_decay = self.norm_decay pattr = ParamAttr(name=(bn_name + '_scale'), learning_rate=norm_lr, regularizer=L2Decay(norm_decay)) battr = ParamAttr(name=(bn_name + '_offset'), learning_rate=norm_lr, regularizer=L2Decay(norm_decay)) if (self.norm_type in ['bn', 'sync_bn']): global_stats = (True if self.freeze_norm else False) out = fluid.layers.batch_norm(input=conv, act=act, name=(bn_name + '.output.1'), param_attr=pattr, bias_attr=battr, moving_mean_name=(bn_name + '_mean'), moving_variance_name=(bn_name + '_variance'), use_global_stats=global_stats) scale = fluid.framework._get_var(pattr.name) bias = fluid.framework._get_var(battr.name) elif (self.norm_type == 'affine_channel'): scale = fluid.layers.create_parameter(shape=[conv.shape[1]], dtype=conv.dtype, attr=pattr, default_initializer=fluid.initializer.Constant(1.0)) bias = fluid.layers.create_parameter(shape=[conv.shape[1]], dtype=conv.dtype, attr=battr, default_initializer=fluid.initializer.Constant(0.0)) out = fluid.layers.affine_channel(x=conv, scale=scale, bias=bias, act=act) if self.freeze_norm: scale.stop_gradient = True bias.stop_gradient = True return out def _shortcut(self, input, ch_out, stride, is_first, name): max_pooling_in_short_cut = (self.variant == 'd') ch_in = input.shape[1] name = self.na.fix_shortcut_name(name) std_senet = getattr(self, 'std_senet', False) if ((ch_in != ch_out) or (stride != 1) or ((self.depth < 50) and is_first)): if std_senet: if is_first: return self._conv_norm(input, ch_out, 1, stride, name=name) else: return self._conv_norm(input, ch_out, 3, stride, name=name) if (max_pooling_in_short_cut and (not is_first)): input = fluid.layers.pool2d(input=input, pool_size=2, pool_stride=2, pool_padding=0, ceil_mode=True, pool_type='avg') return self._conv_norm(input, ch_out, 1, 1, name=name) return self._conv_norm(input, ch_out, 1, stride, name=name) else: return input def bottleneck(self, input, num_filters, stride, is_first, name, dcn_v2=False, gcb=False, gcb_name=None, sac=False, rfp=False, rfp_feat=None, eca_k_size=None): if (self.variant == 'a'): (stride1, stride2) = (stride, 1) else: (stride1, stride2) = (1, stride) groups = getattr(self, 'groups', 1) group_width = getattr(self, 'group_width', (- 1)) if (groups == 1): expand = 4 elif ((groups * group_width) == 256): expand = 1 else: num_filters = (num_filters // 2) expand = 2 (conv_name1, conv_name2, conv_name3, shortcut_name) = self.na.fix_bottleneck_name(name) std_senet = getattr(self, 'std_senet', False) if std_senet: conv_def = [[int((num_filters / 2)), 1, stride1, 'relu', 1, conv_name1], [num_filters, 3, stride2, 'relu', groups, conv_name2], [(num_filters * expand), 1, 1, None, 1, conv_name3]] else: conv_def = [[num_filters, 1, stride1, 'relu', 1, conv_name1], [num_filters, 3, stride2, 'relu', groups, conv_name2], [(num_filters * expand), 1, 1, None, 1, conv_name3]] residual = input for (i, (c, k, s, act, g, _name)) in enumerate(conv_def): residual = self._conv_norm(input=residual, num_filters=c, filter_size=k, stride=s, act=act, groups=g, name=_name, dcn_v2=((i == 1) and dcn_v2), with_sac=((i == 1) and sac)) short = self._shortcut(input, (num_filters * expand), stride, is_first=is_first, name=shortcut_name) if callable(getattr(self, '_squeeze_excitation', None)): residual = self._squeeze_excitation(input=residual, num_channels=num_filters, name=('fc' + name)) if eca_k_size: residual = eca_layer(residual, eca_k_size, name=('eca_' + name)) if gcb: residual = add_gc_block(residual, name=gcb_name, **self.gcb_params) if (rfp and (rfp_feat is not None)): out = (short + residual) rfp_feat = fluid.layers.conv2d(input=rfp_feat, num_filters=(num_filters * expand), filter_size=1, stride=1, padding=0, param_attr=ParamAttr(initializer=Constant(0.0), name=(name + 'rfp.w')), bias_attr=ParamAttr(initializer=Constant(0.0), name=(name + 'rfp.b')), act=None, name=(name + '.rfp.output')) out = fluid.layers.elementwise_add(x=out, y=rfp_feat, act='relu', name=(name + '.add.output.5')) else: out = fluid.layers.elementwise_add(x=short, y=residual, act='relu', name=(name + '.add.output.5')) return out def basicblock(self, input, num_filters, stride, is_first, name, dcn_v2=False, gcb=False, gcb_name=None, sac=False, rfp=False, rfp_feat=None, eca_k_size=None): assert (dcn_v2 is False), 'Not implemented yet.' assert (gcb is False), 'Not implemented yet.' assert (sac is False), 'Not implemented yet.' assert (rfp is False), 'Not implemented yet.' conv0 = self._conv_norm(input=input, num_filters=num_filters, filter_size=3, act='relu', stride=stride, name=(name + '_branch2a')) conv1 = self._conv_norm(input=conv0, num_filters=num_filters, filter_size=3, act=None, name=(name + '_branch2b')) if eca_k_size: conv1 = eca_layer(conv1, eca_k_size, name=('eca_' + name)) short = self._shortcut(input, num_filters, stride, is_first, name=(name + '_branch1')) return fluid.layers.elementwise_add(x=short, y=conv1, act='relu') def layer_warp(self, input, stage_num, rfp_feat=None): assert (stage_num in [2, 3, 4, 5]) self.stage_num = stage_num (stages, block_func) = self.depth_cfg[self.depth] count = stages[(stage_num - 2)] ch_out = self.stage_filters[(stage_num - 2)] is_first = (False if (stage_num != 2) else True) dcn_v2 = (True if (stage_num in self.dcn_v2_stages) else False) sac = (self.with_sac if (stage_num in self.sac_stages) else False) nonlocal_mod = 1000 if (stage_num in self.nonlocal_stages): nonlocal_mod = (self.nonlocal_mod_cfg[self.depth] if (stage_num == 4) else 2) conv = input for i in range(count): conv_name = self.na.fix_layer_warp_name(stage_num, count, i) if (self.depth < 50): is_first = (True if ((i == 0) and (stage_num == 2)) else False) gcb = (stage_num in self.gcb_stages) gcb_name = 'gcb_res{}_b{}'.format(stage_num, i) eca_k_size = (int(self.eca_k_size[(stage_num - 2)]) if self.with_eca else None) rfp = ((i == 0) and (stage_num in self.rfp_stages)) conv = block_func(input=conv, num_filters=ch_out, stride=(2 if ((i == 0) and (stage_num != 2)) else 1), is_first=is_first, name=conv_name, dcn_v2=dcn_v2, gcb=gcb, gcb_name=gcb_name, sac=sac, rfp=rfp, rfp_feat=rfp_feat, eca_k_size=eca_k_size) dim_in = conv.shape[1] nonlocal_name = 'nonlocal_conv{}'.format(stage_num) if ((i % nonlocal_mod) == (nonlocal_mod - 1)): conv = add_space_nonlocal(conv, dim_in, dim_in, (nonlocal_name + '_{}'.format(i)), int((dim_in / 2))) return conv def c1_stage(self, input): out_chan = self._c1_out_chan_num conv1_name = self.na.fix_c1_stage_name() if (self.variant in ['c', 'd']): conv_def = [[(out_chan // 2), 3, 2, 'conv1_1'], [(out_chan // 2), 3, 1, 'conv1_2'], [out_chan, 3, 1, 'conv1_3']] else: conv_def = [[out_chan, 7, 2, conv1_name]] for (c, k, s, _name) in conv_def: input = self._conv_norm(input=input, num_filters=c, filter_size=k, stride=s, act='relu', name=_name) output = fluid.layers.pool2d(input=input, pool_size=3, pool_stride=2, pool_padding=1, pool_type='max') return output def __call__(self, input, rfp_feats=None): assert isinstance(input, Variable) assert (not (set(self.feature_maps) - set([2, 3, 4, 5]))), 'feature maps {} not in [2, 3, 4, 5]'.format(self.feature_maps) res_endpoints = [] res = input feature_maps = self.feature_maps severed_head = getattr(self, 'severed_head', False) if (not severed_head): res = self.c1_stage(res) feature_maps = range(2, (max(self.feature_maps) + 1)) for i in feature_maps: rfp_feat = None if ((i in self.rfp_stages) and (rfp_feats is not None)): rfp_feat = rfp_feats[(i - self.rfp_stages[0])] res = self.layer_warp(res, i, rfp_feat=rfp_feat) if (i in self.feature_maps): res_endpoints.append(res) if (self.freeze_at >= i): res.stop_gradient = True return OrderedDict([('res{}_sum'.format(self.feature_maps[idx]), feat) for (idx, feat) in enumerate(res_endpoints)])
def weights_init_(m): classname = m.__class__.__name__ if (classname.find('Linear') != (- 1)): torch.nn.init.xavier_uniform_(m.weight, gain=1) torch.nn.init.constant_(m.bias, 0)
class ROIHeadsTest(unittest.TestCase): def test_roi_heads(self): torch.manual_seed(121) cfg = get_cfg() cfg.MODEL.ROI_BOX_HEAD.NAME = 'FastRCNNConvFCHead' cfg.MODEL.ROI_BOX_HEAD.NUM_FC = 2 cfg.MODEL.ROI_BOX_HEAD.POOLER_TYPE = 'ROIAlignV2' cfg.MODEL.ROI_BOX_HEAD.BBOX_REG_WEIGHTS = (10, 10, 5, 5) cfg.MODEL.MASK_ON = True num_images = 2 images_tensor = torch.rand(num_images, 20, 30) image_sizes = [(10, 10), (20, 30)] images = ImageList(images_tensor, image_sizes) num_channels = 1024 features = {'res4': torch.rand(num_images, num_channels, 1, 2)} feature_shape = {'res4': ShapeSpec(channels=num_channels, stride=16)} image_shape = (15, 15) gt_boxes0 = torch.tensor([[1, 1, 3, 3], [2, 2, 6, 6]], dtype=torch.float32) gt_instance0 = Instances(image_shape) gt_instance0.gt_boxes = Boxes(gt_boxes0) gt_instance0.gt_classes = torch.tensor([2, 1]) gt_instance0.gt_masks = BitMasks((torch.rand(((2,) + image_shape)) > 0.5)) gt_boxes1 = torch.tensor([[1, 5, 2, 8], [7, 3, 10, 5]], dtype=torch.float32) gt_instance1 = Instances(image_shape) gt_instance1.gt_boxes = Boxes(gt_boxes1) gt_instance1.gt_classes = torch.tensor([1, 2]) gt_instance1.gt_masks = BitMasks((torch.rand(((2,) + image_shape)) > 0.5)) gt_instances = [gt_instance0, gt_instance1] proposal_generator = build_proposal_generator(cfg, feature_shape) roi_heads = StandardROIHeads(cfg, feature_shape) with EventStorage(): (proposals, proposal_losses) = proposal_generator(images, features, gt_instances) (_, detector_losses) = roi_heads(images, features, proposals, gt_instances) detector_losses.update(proposal_losses) expected_losses = {'loss_cls': 4., 'loss_box_reg': 0., 'loss_mask': 0., 'loss_rpn_cls': 0., 'loss_rpn_loc': 0.} succ = all((torch.allclose(detector_losses[name], torch.tensor(expected_losses.get(name, 0.0))) for name in detector_losses.keys())) self.assertTrue(succ, 'Losses has changed! New losses: {}'.format({k: v.item() for (k, v) in detector_losses.items()})) def test_rroi_heads(self): torch.manual_seed(121) cfg = get_cfg() cfg.MODEL.PROPOSAL_GENERATOR.NAME = 'RRPN' cfg.MODEL.ANCHOR_GENERATOR.NAME = 'RotatedAnchorGenerator' cfg.MODEL.ROI_HEADS.NAME = 'RROIHeads' cfg.MODEL.ROI_BOX_HEAD.NAME = 'FastRCNNConvFCHead' cfg.MODEL.ROI_BOX_HEAD.NUM_FC = 2 cfg.MODEL.RPN.BBOX_REG_WEIGHTS = (1, 1, 1, 1, 1) cfg.MODEL.RPN.HEAD_NAME = 'StandardRPNHead' cfg.MODEL.ROI_BOX_HEAD.POOLER_TYPE = 'ROIAlignRotated' cfg.MODEL.ROI_BOX_HEAD.BBOX_REG_WEIGHTS = (10, 10, 5, 5, 1) num_images = 2 images_tensor = torch.rand(num_images, 20, 30) image_sizes = [(10, 10), (20, 30)] images = ImageList(images_tensor, image_sizes) num_channels = 1024 features = {'res4': torch.rand(num_images, num_channels, 1, 2)} feature_shape = {'res4': ShapeSpec(channels=num_channels, stride=16)} image_shape = (15, 15) gt_boxes0 = torch.tensor([[2, 2, 2, 2, 30], [4, 4, 4, 4, 0]], dtype=torch.float32) gt_instance0 = Instances(image_shape) gt_instance0.gt_boxes = RotatedBoxes(gt_boxes0) gt_instance0.gt_classes = torch.tensor([2, 1]) gt_boxes1 = torch.tensor([[1.5, 5.5, 1, 3, 0], [8.5, 4, 3, 2, (- 50)]], dtype=torch.float32) gt_instance1 = Instances(image_shape) gt_instance1.gt_boxes = RotatedBoxes(gt_boxes1) gt_instance1.gt_classes = torch.tensor([1, 2]) gt_instances = [gt_instance0, gt_instance1] proposal_generator = build_proposal_generator(cfg, feature_shape) roi_heads = build_roi_heads(cfg, feature_shape) with EventStorage(): (proposals, proposal_losses) = proposal_generator(images, features, gt_instances) (_, detector_losses) = roi_heads(images, features, proposals, gt_instances) detector_losses.update(proposal_losses) expected_losses = {'loss_cls': 4., 'loss_box_reg': 0., 'loss_rpn_cls': 0., 'loss_rpn_loc': 0.} succ = all((torch.allclose(detector_losses[name], torch.tensor(expected_losses.get(name, 0.0))) for name in detector_losses.keys())) self.assertTrue(succ, 'Losses has changed! New losses: {}'.format({k: v.item() for (k, v) in detector_losses.items()})) def test_box_head_scriptability(self): input_shape = ShapeSpec(channels=1024, height=14, width=14) box_features = torch.randn(4, 1024, 14, 14) box_head = FastRCNNConvFCHead(input_shape, conv_dims=[512, 512], fc_dims=[1024, 1024]).eval() script_box_head = torch.jit.script(box_head) origin_output = box_head(box_features) script_output = script_box_head(box_features) self.assertTrue(torch.equal(origin_output, script_output)) def test_mask_head_scriptability(self): input_shape = ShapeSpec(channels=1024) mask_features = torch.randn(4, 1024, 14, 14) image_shapes = [(10, 10), (15, 15)] pred_instance0 = Instances(image_shapes[0]) pred_classes0 = torch.tensor([1, 2, 3], dtype=torch.int64) pred_instance0.pred_classes = pred_classes0 pred_instance1 = Instances(image_shapes[1]) pred_classes1 = torch.tensor([4], dtype=torch.int64) pred_instance1.pred_classes = pred_classes1 mask_head = MaskRCNNConvUpsampleHead(input_shape, num_classes=80, conv_dims=[256, 256]).eval() origin_outputs = mask_head(mask_features, deepcopy([pred_instance0, pred_instance1])) fields = {'pred_masks': torch.Tensor, 'pred_classes': torch.Tensor} with freeze_training_mode(mask_head), patch_instances(fields) as NewInstances: sciript_mask_head = torch.jit.script(mask_head) pred_instance0 = NewInstances.from_instances(pred_instance0) pred_instance1 = NewInstances.from_instances(pred_instance1) script_outputs = sciript_mask_head(mask_features, [pred_instance0, pred_instance1]) for (origin_ins, script_ins) in zip(origin_outputs, script_outputs): assert_instances_allclose(origin_ins, script_ins, rtol=0) def test_keypoint_head_scriptability(self): input_shape = ShapeSpec(channels=1024, height=14, width=14) keypoint_features = torch.randn(4, 1024, 14, 14) image_shapes = [(10, 10), (15, 15)] pred_boxes0 = torch.tensor([[1, 1, 3, 3], [2, 2, 6, 6], [1, 5, 2, 8]], dtype=torch.float32) pred_instance0 = Instances(image_shapes[0]) pred_instance0.pred_boxes = Boxes(pred_boxes0) pred_boxes1 = torch.tensor([[7, 3, 10, 5]], dtype=torch.float32) pred_instance1 = Instances(image_shapes[1]) pred_instance1.pred_boxes = Boxes(pred_boxes1) keypoint_head = KRCNNConvDeconvUpsampleHead(input_shape, num_keypoints=17, conv_dims=[512, 512]).eval() origin_outputs = keypoint_head(keypoint_features, deepcopy([pred_instance0, pred_instance1])) fields = {'pred_boxes': Boxes, 'pred_keypoints': torch.Tensor, 'pred_keypoint_heatmaps': torch.Tensor} with freeze_training_mode(keypoint_head), patch_instances(fields) as NewInstances: script_keypoint_head = torch.jit.script(keypoint_head) pred_instance0 = NewInstances.from_instances(pred_instance0) pred_instance1 = NewInstances.from_instances(pred_instance1) script_outputs = script_keypoint_head(keypoint_features, [pred_instance0, pred_instance1]) for (origin_ins, script_ins) in zip(origin_outputs, script_outputs): assert_instances_allclose(origin_ins, script_ins, rtol=0) def test_StandardROIHeads_scriptability(self): cfg = get_cfg() cfg.MODEL.ROI_BOX_HEAD.NAME = 'FastRCNNConvFCHead' cfg.MODEL.ROI_BOX_HEAD.NUM_FC = 2 cfg.MODEL.ROI_BOX_HEAD.POOLER_TYPE = 'ROIAlignV2' cfg.MODEL.ROI_BOX_HEAD.BBOX_REG_WEIGHTS = (10, 10, 5, 5) cfg.MODEL.MASK_ON = True cfg.MODEL.ROI_HEADS.NMS_THRESH_TEST = 0.01 cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.01 num_images = 2 images_tensor = torch.rand(num_images, 20, 30) image_sizes = [(10, 10), (20, 30)] images = ImageList(images_tensor, image_sizes) num_channels = 1024 features = {'res4': torch.rand(num_images, num_channels, 1, 2)} feature_shape = {'res4': ShapeSpec(channels=num_channels, stride=16)} roi_heads = StandardROIHeads(cfg, feature_shape).eval() proposal0 = Instances(image_sizes[0]) proposal_boxes0 = torch.tensor([[1, 1, 3, 3], [2, 2, 6, 6]], dtype=torch.float32) proposal0.proposal_boxes = Boxes(proposal_boxes0) proposal0.objectness_logits = torch.tensor([0.5, 0.7], dtype=torch.float32) proposal1 = Instances(image_sizes[1]) proposal_boxes1 = torch.tensor([[1, 5, 2, 8], [7, 3, 10, 5]], dtype=torch.float32) proposal1.proposal_boxes = Boxes(proposal_boxes1) proposal1.objectness_logits = torch.tensor([0.1, 0.9], dtype=torch.float32) proposals = [proposal0, proposal1] (pred_instances, _) = roi_heads(images, features, proposals) fields = {'objectness_logits': torch.Tensor, 'proposal_boxes': Boxes, 'pred_classes': torch.Tensor, 'scores': torch.Tensor, 'pred_masks': torch.Tensor, 'pred_boxes': Boxes, 'pred_keypoints': torch.Tensor, 'pred_keypoint_heatmaps': torch.Tensor} with freeze_training_mode(roi_heads), patch_instances(fields) as new_instances: proposal0 = new_instances.from_instances(proposal0) proposal1 = new_instances.from_instances(proposal1) proposals = [proposal0, proposal1] scripted_rot_heads = torch.jit.script(roi_heads) (scripted_pred_instances, _) = scripted_rot_heads(images, features, proposals) for (instance, scripted_instance) in zip(pred_instances, scripted_pred_instances): assert_instances_allclose(instance, scripted_instance, rtol=0) def test_PointRend_mask_head_tracing(self): cfg = model_zoo.get_config('COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yaml') point_rend.add_pointrend_config(cfg) cfg.MODEL.ROI_HEADS.IN_FEATURES = ['p2', 'p3'] cfg.MODEL.ROI_MASK_HEAD.NAME = 'PointRendMaskHead' cfg.MODEL.ROI_MASK_HEAD.POOLER_TYPE = '' cfg.MODEL.ROI_MASK_HEAD.POINT_HEAD_ON = True chan = 256 head = point_rend.PointRendMaskHead(cfg, {'p2': ShapeSpec(channels=chan, stride=4), 'p3': ShapeSpec(channels=chan, stride=8)}) def gen_inputs(h, w, N): p2 = torch.rand(1, chan, h, w) p3 = torch.rand(1, chan, (h // 2), (w // 2)) boxes = random_boxes(N, max_coord=h) return (p2, p3, boxes) class Wrap(nn.ModuleDict): def forward(self, p2, p3, boxes): features = {'p2': p2, 'p3': p3} inst = Instances(((p2.shape[2] * 4), (p2.shape[3] * 4))) inst.pred_boxes = Boxes(boxes) inst.pred_classes = torch.zeros(inst.__len__(), dtype=torch.long) out = self.head(features, [inst])[0] return out.pred_masks model = Wrap({'head': head}) model.eval() with torch.no_grad(), patch_builtin_len(): traced = torch.jit.trace(model, gen_inputs(302, 208, 20)) inputs = gen_inputs(100, 120, 30) out_eager = model(*inputs) out_trace = traced(*inputs) self.assertTrue(torch.allclose(out_eager, out_trace))
def load_flax_weights_in_pytorch_model(pt_model, flax_state): try: import torch except ImportError: logger.error('Loading Flax weights in PyTorch requires both PyTorch and Flax to be installed. Please see and for installation instructions.') raise is_type_bf16 = flatten_dict(jax.tree_util.tree_map((lambda x: (x.dtype == jnp.bfloat16)), flax_state)).values() if any(is_type_bf16): logger.warning('Found ``bfloat16`` weights in Flax model. Casting all ``bfloat16`` weights to ``float32`` before loading those in PyTorch model.') flax_state = jax.tree_util.tree_map((lambda params: (params.astype(np.float32) if (params.dtype == jnp.bfloat16) else params)), flax_state) pt_model.base_model_prefix = '' flax_state_dict = flatten_dict(flax_state, sep='.') pt_model_dict = pt_model.state_dict() unexpected_keys = [] missing_keys = set(pt_model_dict.keys()) for (flax_key_tuple, flax_tensor) in flax_state_dict.items(): flax_key_tuple_array = flax_key_tuple.split('.') if ((flax_key_tuple_array[(- 1)] == 'kernel') and (flax_tensor.ndim == 4)): flax_key_tuple_array = (flax_key_tuple_array[:(- 1)] + ['weight']) flax_tensor = jnp.transpose(flax_tensor, (3, 2, 0, 1)) elif (flax_key_tuple_array[(- 1)] == 'kernel'): flax_key_tuple_array = (flax_key_tuple_array[:(- 1)] + ['weight']) flax_tensor = flax_tensor.T elif (flax_key_tuple_array[(- 1)] == 'scale'): flax_key_tuple_array = (flax_key_tuple_array[:(- 1)] + ['weight']) if ('time_embedding' not in flax_key_tuple_array): for (i, flax_key_tuple_string) in enumerate(flax_key_tuple_array): flax_key_tuple_array[i] = flax_key_tuple_string.replace('_0', '.0').replace('_1', '.1').replace('_2', '.2').replace('_3', '.3').replace('_4', '.4').replace('_5', '.5').replace('_6', '.6').replace('_7', '.7').replace('_8', '.8').replace('_9', '.9') flax_key = '.'.join(flax_key_tuple_array) if (flax_key in pt_model_dict): if (flax_tensor.shape != pt_model_dict[flax_key].shape): raise ValueError(f'Flax checkpoint seems to be incorrect. Weight {flax_key_tuple} was expected to be of shape {pt_model_dict[flax_key].shape}, but is {flax_tensor.shape}.') else: flax_tensor = (np.asarray(flax_tensor) if (not isinstance(flax_tensor, np.ndarray)) else flax_tensor) pt_model_dict[flax_key] = torch.from_numpy(flax_tensor) missing_keys.remove(flax_key) else: unexpected_keys.append(flax_key) pt_model.load_state_dict(pt_model_dict) missing_keys = list(missing_keys) if (len(unexpected_keys) > 0): logger.warning(f'''Some weights of the Flax model were not used when initializing the PyTorch model {pt_model.__class__.__name__}: {unexpected_keys} - This IS expected if you are initializing {pt_model.__class__.__name__} from a Flax model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a FlaxBertForPreTraining model). - This IS NOT expected if you are initializing {pt_model.__class__.__name__} from a Flax model that you expect to be exactly identical (e.g. initializing a BertForSequenceClassification model from a FlaxBertForSequenceClassification model).''') if (len(missing_keys) > 0): logger.warning(f'''Some weights of {pt_model.__class__.__name__} were not initialized from the Flax model and are newly initialized: {missing_keys} You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.''') return pt_model
class LMTrainer(Trainer): def __init__(self, opt, model, data): super().__init__(opt, model, data) self.crit = nn.CrossEntropyLoss(size_average=True, reduce=True, ignore_index=0) self.crit_test = nn.CrossEntropyLoss(size_average=False, reduce=False, ignore_index=0) self.opt = opt self.model = model self.train_bag = data[0] self.test_bag = data[1] self.n_batch = len(self.train_bag) self.optimizer = torch.optim.SGD(self.model.parameters(), lr=opt.lr, momentum=0.9) self.word_dict = opt.word_dict self.clip = opt.clip def func_train(self, inp_var, inp_msk): self.optimizer.zero_grad() aux = {} batch_size = inp_var.size()[0] batch_size_ = len(inp_msk) assert (batch_size == batch_size_) target_len = inp_msk[0] (decoder_outputs_prob, decoder_outputs) = self.model.forward(inp_var, inp_msk, inp_var, inp_msk, aux) valid_pos_mask = Var(msk_list_to_mat(inp_msk), requires_grad=False).view((target_len * batch_size), 1) if self.opt.use_cuda: valid_pos_mask = valid_pos_mask.cuda() pred_prob = decoder_outputs_prob.view((target_len * batch_size), (- 1)) seq_first_inp_var = inp_var.transpose(1, 0).contiguous() gold_dist = seq_first_inp_var.view((target_len * batch_size)) gold_dist = Var(gold_dist) if self.opt.use_cuda: gold_dist = gold_dist.cuda() loss = self.crit(pred_prob, gold_dist) loss.backward() torch.nn.utils.clip_grad_norm(self.model.parameters(), self.clip) self.optimizer.step() return (loss.data[0], math.exp(loss.data[0])) def func_test(self, inp_var, inp_msk): target_len = inp_msk[0] batch_size = inp_var.size()[0] (decoder_outputs_prob, decoder_outputs) = self.model.forward(inp_var, inp_msk, tgt_var=inp_var, tgt_msk=inp_msk, aux=None) pred_prob = decoder_outputs_prob.view((target_len * batch_size), (- 1)) seq_first_inp_var = inp_var.transpose(1, 0).contiguous() gold_dist = Var(seq_first_inp_var.view((target_len * batch_size))) if self.opt.use_cuda: gold_dist = gold_dist.cuda() loss = self.crit_test(pred_prob, gold_dist) loss = torch.sum(loss) return (loss.data[0], decoder_outputs) def train_iters(self): for epo in range(self.opt.start_epo, (self.opt.n_epo + 1)): self.model.train() batch_order = np.arange(self.n_batch) np.random.shuffle(batch_order) for (idx, batch_idx) in enumerate(batch_order): current_batch = self.train_bag[batch_idx] inp_var = current_batch['txt'] inp_msk = current_batch['txt_msk'] if self.opt.use_cuda: inp_var = inp_var.contiguous().cuda() (nll, ppl) = self.func_train(inp_var, inp_msk) if ((idx % self.opt.print_every) == 0): logging.info(('NLL:%.2f \tPPL:%s' % (nll, str(ppl)))) if ((idx % self.opt.save_every) == 0): ppl = self.evaluate() os.chdir(self.opt.save_dir) name_string = ('%d_%.2f'.lower() % (epo, ppl)) logging.info(('Saving in epo %s' % name_string)) torch.save(self.model.emb.state_dict(), (name_string + '_emb')) torch.save(self.model.dec.state_dict(), (name_string + '_dec')) torch.save(self.model.opt, (name_string + '_opt')) os.chdir('..') def evaluate(self): self.model.eval() n_batch = len(self.test_bag) test_len = 0 accumulated_ppl = 0 for idx in range(n_batch): current_batch = self.test_bag[idx] inp_var = current_batch['txt'] inp_mask = current_batch['txt_msk'] batch_size = inp_var.size()[0] test_len += (inp_mask[0] * batch_size) (nll, decoder_output) = self.func_test(inp_var, inp_mask) accumulated_ppl += nll final_ppl = (accumulated_ppl / test_len) final_ppl = math.exp(final_ppl) logging.info(('PPL: %f' % final_ppl)) return final_ppl
class _DefaultHabitatSimActions(Enum): STOP = 0 MOVE_FORWARD = 1 TURN_LEFT = 2 TURN_RIGHT = 3 LOOK_UP = 4 LOOK_DOWN = 5
class SquadFeatures(object): def __init__(self, input_ids, attention_mask, token_type_ids, cls_index, p_mask, example_index, unique_id, paragraph_len, token_is_max_context, tokens, token_to_orig_map, start_position, end_position, is_impossible): self.input_ids = input_ids self.attention_mask = attention_mask self.token_type_ids = token_type_ids self.cls_index = cls_index self.p_mask = p_mask self.example_index = example_index self.unique_id = unique_id self.paragraph_len = paragraph_len self.token_is_max_context = token_is_max_context self.tokens = tokens self.token_to_orig_map = token_to_orig_map self.start_position = start_position self.end_position = end_position self.is_impossible = is_impossible
class DiscLossWGANGP(DiscLossLS): def name(self): return 'DiscLossWGAN-GP' def __init__(self, opt, tensor): super(DiscLossWGANGP, self).__init__(opt, tensor) self.LAMBDA = 10 def get_g_loss(self, net, realA, fakeB): self.D_fake = net.forward(fakeB) return (- self.D_fake.mean()) def calc_gradient_penalty(self, netD, real_data, fake_data): alpha = torch.rand(1, 1) alpha = alpha.expand(real_data.size()) alpha = alpha.cuda() interpolates = ((alpha * real_data) + ((1 - alpha) * fake_data)) interpolates = interpolates.cuda() interpolates = Variable(interpolates, requires_grad=True) disc_interpolates = netD.forward(interpolates) gradients = autograd.grad(outputs=disc_interpolates, inputs=interpolates, grad_outputs=torch.ones(disc_interpolates.size()).cuda(), create_graph=True, retain_graph=True, only_inputs=True)[0] gradient_penalty = (((gradients.norm(2, dim=1) - 1) ** 2).mean() * self.LAMBDA) return gradient_penalty def get_loss(self, net, realA, fakeB, realB): self.D_fake = net.forward(fakeB.detach()) self.D_fake = self.D_fake.mean() self.D_real = net.forward(realB) self.D_real = self.D_real.mean() self.loss_D = (self.D_fake - self.D_real) gradient_penalty = self.calc_gradient_penalty(net, realB.data, fakeB.data) return (self.loss_D + gradient_penalty)
def main(): parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments, AdapterTrainingArguments)) if ((len(sys.argv) == 2) and sys.argv[1].endswith('.json')): (model_args, data_args, training_args, adapter_args) = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1])) else: (model_args, data_args, training_args, adapter_args) = parser.parse_args_into_dataclasses() training_args.predict_with_generate = True wandb.init(entity='lklab_kaist', project='ROE_experiments_ICLR', name=training_args.output_dir) last_checkpoint = None if (os.path.isdir(training_args.output_dir) and training_args.do_train and (not training_args.overwrite_output_dir)): last_checkpoint = get_last_checkpoint(training_args.output_dir) print('#### last_checkpoint ', last_checkpoint) if ((last_checkpoint is None) and (len(os.listdir(training_args.output_dir)) > 0)): pass elif (last_checkpoint is not None): logger.info(f'Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change the `--output_dir` or add `--overwrite_output_dir` to train from scratch.') logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s', datefmt='%m/%d/%Y %H:%M:%S', handlers=[logging.StreamHandler(sys.stdout)]) logger.setLevel((logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)) logger.warning((f'Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}' + f'distributed training: {bool((training_args.local_rank != (- 1)))}, 16-bits training: {training_args.fp16}')) if is_main_process(training_args.local_rank): transformers.utils.logging.set_verbosity_info() logger.info('Training/evaluation parameters %s', training_args) set_seed(training_args.seed) config = T5Config.from_pretrained((model_args.config_name if model_args.config_name else model_args.model_name_or_path), cache_dir=model_args.cache_dir, revision=model_args.model_revision, use_auth_token=(True if model_args.use_auth_token else None)) config.train_task_adapters = adapter_args.train_task_adapters config.prefix_tuning = adapter_args.prefix_tuning config.attn_prefix_tuning = model_args.attn_prefix_tuning config.attn_method = model_args.attn_method config.ignore_target = model_args.ignore_target config.shared_attn = model_args.shared_attn config.prefix_num = model_args.prefix_num config.num_target = len(data_args.task_name) config.temperature = model_args.temperature config.fix_attention = model_args.fix_attention adapter_config = get_adapter_config(adapter_args, data_args, training_args, config) tokenizer = AutoTokenizer.from_pretrained((model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path), cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer, revision=model_args.model_revision, use_auth_token=(True if model_args.use_auth_token else None)) model = T5ForConditionalGeneration.from_pretrained(model_args.model_name_or_path, from_tf=bool(('.ckpt' in model_args.model_name_or_path)), config=config, cache_dir=model_args.cache_dir, revision=model_args.model_revision, use_auth_token=(True if model_args.use_auth_token else None), adapter_config=adapter_config) if (model_args.load_prefix_embeddings is True): if (model_args.prompt_embedding_path is None): for (name, param) in model.named_parameters(): if (('prefix_shared' in name) or ('prefix' in name)): shared_params = [param] else: shared_params = [] for path in model_args.prompt_embedding_path: shared_param = torch.load(path) shared_params.append(shared_param) if (model_args.target_prompt_embedding_path is not None): target_prompt_embedding = torch.load(model_args.target_prompt_embedding_path) if (model_args.attn_prefix_tuning is True): if ((training_args.do_train is True) and (model_args.shared_attn is False)): model.store_prefix_weights(shared_params) model.update_prefix_weights_single(shared_params[0]) elif ((training_args.do_train is True) and (model_args.shared_attn is True)): model.store_prefix_weights(shared_params) model.update_prefix_weights_multi(shared_params[0], num_target=config.num_target) else: model.store_prefix_weights(shared_params) model.update_prefix_weights_single(target_prompt_embedding) elif (model_args.target_prompt_embedding_path is None): model.update_prefix_weights(shared_params) else: model.update_prefix_weights(shared_params, target_prompt_embedding) if ((model_args.load_attention is True) and (model_args.attn_path is not None)): model.update_attention_weights(torch.load(model_args.attn_path)) if ((model_args.load_attention is True) and (model_args.attn_path_sub is not None)): model.update_attention_weights_sub(model_args.attn_path_sub) if ((model_args.load_layer_norm is True) and (model_args.layer_norm_dir is not None)): model.update_layer_norm_weights(model_args.layer_norm_dir) model.resize_token_embeddings(len(tokenizer)) model = modify_model_after_init(model, training_args, adapter_args, adapter_config) model_args.load_adapter_weights = True if model_args.load_adapter_weights: adapter_params = {} lst = os.listdir(os.path.join(training_args.output_dir, 'adapter_params')) for path in lst: full_path = os.path.join(training_args.output_dir, 'adapter_params', path) params = torch.load(full_path) path_ = path.split('.') path = '.'.join(path_[:(- 1)]) adapter_params[path] = params load_cnt = 0 for (name, param) in model.named_parameters(): if param.requires_grad: load_cnt += 1 param.data = adapter_params[name].cuda() print(f'load count: {load_cnt}') print(f'Finished loading {len(adapter_params)} number of adapter parameter files') data_args.dataset_name = data_args.task_name data_args.eval_dataset_name = data_args.task_name data_args.test_dataset_name = data_args.test_dataset_name data_args.dataset_config_name = data_args.dataset_config_name data_args.eval_dataset_config_name = data_args.dataset_config_name data_args.test_dataset_config_name = data_args.test_dataset_config_name data_args.eval_prompts = data_args.train_prompts assert (len(data_args.dataset_name) == len(data_args.dataset_config_name)) if (data_args.eval_dataset_name is not None): assert (len(data_args.eval_dataset_name) == len(data_args.eval_dataset_config_name)) if (data_args.test_dataset_name is not None): assert (len(data_args.test_dataset_name) == len(data_args.test_dataset_config_name)) padding = ('max_length' if data_args.pad_to_max_length else False) def preprocess_function(examples, max_target_length, task_id=None): model_inputs = tokenizer(examples['source'], max_length=data_args.max_source_length, padding=padding, truncation=True) with tokenizer.as_target_tokenizer(): labels = tokenizer(examples['target'], max_length=max_target_length, padding=padding, truncation=True) if ((padding == 'max_length') and data_args.ignore_pad_token_for_loss): labels['input_ids'] = [[(l if (l != tokenizer.pad_token_id) else (- 100)) for l in label] for label in labels['input_ids']] model_inputs['labels'] = labels['input_ids'] model_inputs['extra_fields'] = examples['extra_fields'] if (task_id is not None): model_inputs['task_ids'] = [task_id for _ in examples['extra_fields']] return model_inputs column_names = ['source', 'target', 'extra_fields'] performance_metrics = {} eval_metrics_dict = {((dataset_name + '*') + eval_prompt): AutoTask.get(dataset_name, dataset_config_name, prompt=eval_prompt).metric for (dataset_name, dataset_config_name, eval_prompt) in zip(data_args.eval_dataset_name, data_args.eval_dataset_config_name, data_args.eval_prompts)} print('') print(data_args.eval_dataset_name) print() print(data_args.eval_dataset_config_name) print() print(eval_metrics_dict) print('') training_args.do_train = False if training_args.do_train: if (data_args.train_files is not None): train_datasets = [AutoTask.get(dataset_name, dataset_config_name, prompt=train_prompt, seed=data_args.data_seed).get(split='train', split_validation_test=training_args.split_validation_test, add_prefix=(False if adapter_args.train_task_adapters else True), n_obs=data_args.max_train_samples, lang=data_args.lang_name, file_name=train_file) for (dataset_name, dataset_config_name, train_file, train_prompt) in zip(data_args.dataset_name, data_args.dataset_config_name, data_args.train_files, data_args.train_prompts)] for td in train_datasets: print('') print(len(td)) print('') else: train_datasets = [AutoTask.get(dataset_name, dataset_config_name, prompt=train_prompt, seed=data_args.data_seed).get(split='train', split_validation_test=training_args.split_validation_test, add_prefix=(False if adapter_args.train_task_adapters else True), n_obs=data_args.max_train_samples, lang=data_args.lang_name, file_name=data_args.train_file) for (dataset_name, dataset_config_name, train_prompt) in zip(data_args.dataset_name, data_args.dataset_config_name, data_args.train_prompts)] for td in train_datasets: print('') print(len(td)) print('') max_target_lengths = [AutoTask.get(dataset_name, dataset_config_name, prompt=train_prompt).get_max_target_length(tokenizer=tokenizer, default_max_length=data_args.max_target_length) for (dataset_name, dataset_config_name, train_prompt) in zip(data_args.dataset_name, data_args.dataset_config_name, data_args.train_prompts)] for (i, train_dataset) in enumerate(train_datasets): if (model_args.shared_attn is True): train_datasets[i] = train_datasets[i].map(functools.partial(preprocess_function, max_target_length=max_target_lengths[i], task_id=i), batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=(not data_args.overwrite_cache)) else: print('') print(len(train_datasets[i])) print('') train_datasets[i] = train_datasets[i].map(functools.partial(preprocess_function, max_target_length=max_target_lengths[i]), batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=(not data_args.overwrite_cache)) print('') print(len(train_datasets[i])) print('') print('') print(len(train_dataset)) print('') train_dataset = concatenate_datasets(train_datasets) print('') print(len(train_dataset)) print('') training_args.do_eval = True data_args.max_val_samples = 300 if training_args.do_eval: max_target_lengths = [AutoTask.get(dataset_name, dataset_config_name, prompt=eval_prompt).get_max_target_length(tokenizer=tokenizer, default_max_length=data_args.max_target_length) for (dataset_name, dataset_config_name, eval_prompt) in zip(data_args.eval_dataset_name, data_args.eval_dataset_config_name, data_args.eval_prompts)] if (data_args.validation_files is not None): eval_datasets = {((eval_dataset + '*') + eval_prompt): AutoTask.get(eval_dataset, eval_dataset_config, prompt=eval_prompt, seed=data_args.data_seed).get(split='validation', split_validation_test=training_args.split_validation_test, add_prefix=(False if adapter_args.train_task_adapters else True), n_obs=data_args.max_val_samples, lang=data_args.lang_name, file_name=validation_file) for (eval_dataset, eval_dataset_config, validation_file, eval_prompt) in zip(data_args.eval_dataset_name, data_args.eval_dataset_config_name, data_args.validation_files, data_args.eval_prompts)} else: eval_datasets = {((eval_dataset + '*') + eval_prompt): AutoTask.get(eval_dataset, eval_dataset_config, prompt=eval_prompt, seed=data_args.data_seed).get(split='validation', split_validation_test=training_args.split_validation_test, add_prefix=(False if adapter_args.train_task_adapters else True), n_obs=data_args.max_val_samples, lang=data_args.lang_name, file_name=data_args.validation_file) for (eval_dataset, eval_dataset_config, eval_prompt) in zip(data_args.eval_dataset_name, data_args.eval_dataset_config_name, data_args.eval_prompts)} for (k, name) in enumerate(eval_datasets): if (name == 'lama_fill_mask'): max_target_lengths[k] = 2 elif (name == 'lambada_what comes next'): max_target_lengths[k] = 1 if (model_args.shared_attn is True): eval_datasets[name] = eval_datasets[name].map(functools.partial(preprocess_function, max_target_length=max_target_lengths[k], task_id=k), batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=(not data_args.overwrite_cache)) else: eval_datasets[name] = eval_datasets[name].map(functools.partial(preprocess_function, max_target_length=max_target_lengths[k]), batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=(not data_args.overwrite_cache)) if training_args.do_test: if (data_args.test_files is not None): test_datasets = {test_dataset: AutoTask.get(test_dataset, test_dataset_config, prompt=test_prompt, seed=data_args.data_seed).get(split='test', split_validation_test=training_args.split_validation_test, add_prefix=(False if adapter_args.train_task_adapters else True), n_obs=data_args.max_test_samples, lang=data_args.lang_name, file_name=test_file) for (test_dataset, test_dataset_config, test_file, test_prompt) in zip(data_args.test_dataset_name, data_args.test_dataset_config_name, data_args.test_files, data_args.test_prompts)} else: test_datasets = {test_dataset: AutoTask.get(test_dataset, test_dataset_config, prompt=test_prompt, seed=data_args.data_seed).get(split='test', split_validation_test=training_args.split_validation_test, add_prefix=(False if adapter_args.train_task_adapters else True), n_obs=data_args.max_test_samples, lang=data_args.lang_name, file_name=data_args.test_file) for (test_dataset, test_dataset_config, test_prompt) in zip(data_args.test_dataset_name, data_args.test_dataset_config_name, data_args.test_prompts)} max_target_lengths = [AutoTask.get(dataset_name, dataset_config_name, prompt=test_prompt).get_max_target_length(tokenizer=tokenizer, default_max_length=data_args.max_target_length) for (dataset_name, dataset_config_name, test_prompt) in zip(data_args.test_dataset_name, data_args.test_dataset_config_name, data_args.test_prompts)] for (k, name) in enumerate(test_datasets): if (name == 'lama'): max_target_lengths[k] = 2 elif (name == 'lambada'): max_target_lengths[k] = 1 if (model_args.shared_attn is True): test_datasets[name] = test_datasets[name].map(functools.partial(preprocess_function, max_target_length=max_target_lengths[k], task_id=k), batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=(not data_args.overwrite_cache)) else: test_datasets[name] = test_datasets[name].map(functools.partial(preprocess_function, max_target_length=max_target_lengths[k]), batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=(not data_args.overwrite_cache)) label_pad_token_id = ((- 100) if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id) if data_args.pad_to_max_length: data_collator = default_data_collator else: data_collator = TaskDataCollatorForSeq2Seq(tokenizer, label_pad_token_id=label_pad_token_id, pad_to_multiple_of=(8 if training_args.fp16 else None)) if training_args.do_eval: data_info = {'eval': eval_datasets[((data_args.eval_dataset_name[0] + '*') + data_args.eval_prompts[0])]['extra_fields'], 'test': (test_datasets[((data_args.test_dataset_name[0] + '*') + data_args.test_prompts[0])]['extra_fields'] if training_args.do_test else None), 'train': (train_dataset['extra_fields'] if training_args.do_train else None)} else: data_info = {'train': (train_dataset['extra_fields'] if training_args.do_train else None)} def compute_metrics(eval_preds, task_name): print('') print(eval_preds) print('') (preds, labels, data_info) = eval_preds post_processor = AutoPostProcessor.get(task_name, tokenizer, data_args.ignore_pad_token_for_loss) (decoded_preds, decoded_labels) = post_processor.process(preds, labels, data_info) result = {} eval_metrics = eval_metrics_dict[task_name] for metric in eval_metrics: result.update(metric(decoded_preds, decoded_labels)) with open(f"./output_logs_verbalizer/{data_args.dataset_name[0].replace('/', ' ').replace('-', ' ')}*{data_args.train_prompts[0].replace('/', ' ').replace('-', ' ')}-{task_name.replace('/', ' ').replace('-', ' ')}.txt", 'a') as f: f.write('\n') f.write(task_name) f.write('\n') for (a, b) in zip(decoded_preds, decoded_labels): f.write(a) f.write(' ') f.write(b) f.write('\n') f.write('>> ') for (key, value) in result.items(): f.write((((str(key) + ' : ') + str(value)) + ' | ')) f.write('\n') f.write('\n') return result if (model_args.attn_learning_rate is not None): all_parameters = set(model.parameters()) attn_params = [] for (name, param) in model.named_parameters(): if ((name == 'encoder.attn_W_up') or (name == 'encoder.attn_W_down') or (name == 'encoder.layer_norm')): attn_params += list(param) attn_params = set(attn_params) non_attn_params = (all_parameters - attn_params) non_attn_params = list(non_attn_params) attn_params = list(attn_params) optim = AdamW([{'params': non_attn_params}, {'params': attn_params, 'lr': model_args.attn_learning_rate}], lr=training_args.learning_rate) scheduler = get_linear_schedule_with_warmup(optim, num_warmup_steps=training_args.warmup_steps, num_training_steps=((len(train_dataset) * training_args.num_train_epochs) // (training_args.gradient_accumulation_steps * training_args.per_device_train_batch_size))) trainer = Seq2SeqTrainer(model=model, args=training_args, data_args=data_args, train_dataset=(train_dataset if training_args.do_train else None), eval_datasets=(eval_datasets if training_args.do_eval else None), data_info=data_info, tokenizer=tokenizer, data_collator=data_collator, compute_metrics=compute_metrics, evaluation_metrics=eval_metrics_dict, shared=model_args.shared_attn, optimizers=(optim, scheduler)) else: trainer = Seq2SeqTrainer(model=model, args=training_args, data_args=data_args, train_dataset=(train_dataset if training_args.do_train else None), eval_datasets=(eval_datasets if training_args.do_eval else None), data_info=data_info, tokenizer=tokenizer, data_collator=data_collator, compute_metrics=compute_metrics, evaluation_metrics=eval_metrics_dict, shared=model_args.shared_attn) if training_args.do_eval: print('') print(eval_datasets) print('') if trainer.is_world_process_zero(): os.makedirs(training_args.output_dir, exist_ok=True) save_training_config(sys.argv[1], training_args.output_dir) model_args.save_adapter_weights = False if model_args.save_adapter_weights: params_to_save = {} unfrozen_layers = 0 for (name, param) in trainer.model.named_parameters(): if (param.requires_grad == True): print(name) params_to_save[name] = 0 unfrozen_layers += 1 print(f'number of unfrozen layers (for beginning of training model): {unfrozen_layers}') if training_args.do_train: checkpoint = None if (training_args.resume_from_checkpoint is not None): checkpoint = training_args.resume_from_checkpoint elif (last_checkpoint is not None): checkpoint = last_checkpoint if training_args.compute_time: torch.cuda.synchronize() start = torch.cuda.Event(enable_timing=True) end = torch.cuda.Event(enable_timing=True) start.record() train_result = trainer.train(resume_from_checkpoint=checkpoint) if training_args.compute_time: end.record() torch.cuda.synchronize() total_time = (start.elapsed_time(end) / (1000 * 60)) performance_metrics.update({'total_time in minutes ': total_time}) if model_args.save_adapter_weights: if (not os.path.exists(os.path.join(training_args.output_dir, 'adapter_params'))): os.mkdir(os.path.join(training_args.output_dir, 'adapter_params')) layer_cnt = 0 for (name, param) in trainer.model.named_parameters(): if (name in params_to_save): save_path = os.path.join(training_args.output_dir, f'adapter_params/{name}.pt') print(name) torch.save(param, save_path) layer_cnt += 1 print(f'finished saving adapters! saved {layer_cnt} number of layers') exit() if model_args.save_prefix_only: for (name, param) in trainer.model.named_parameters(): if ((model_args.attn_prefix_tuning is False) and (('prefix_shared' in name) or ('prefix' in name))): shared_params = param torch.save(shared_params, os.path.join(training_args.output_dir, 'prefix_embeddings.pt')) elif ((model_args.attn_prefix_tuning is True) and (name == 'prefix_shared')): shared_params = param if (model_args.shared_attn is True): for i in range(config.num_target): torch.save(shared_params[i], os.path.join(training_args.output_dir, 'prefix_embeddings_{}.pt'.format(i))) else: torch.save(shared_params, os.path.join(training_args.output_dir, 'prefix_embeddings.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.attn_Wa.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(training_args.output_dir, 'attn_Wa_weights.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.attn_W_down.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(training_args.output_dir, 'attn_W_down.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.attn_W_up.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(training_args.output_dir, 'attn_W_up.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.layer_norm.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(training_args.output_dir, 'layer_norm_weight.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.layer_norm.bias' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(training_args.output_dir, 'layer_norm_bias.pt')) else: trainer.save_model() train_metrics = train_result.metrics max_train_samples = (data_args.max_train_samples if (data_args.max_train_samples is not None) else len(train_dataset)) train_metrics['train_samples'] = min(max_train_samples, len(train_dataset)) trainer.log_metrics('train', train_metrics) trainer.save_metrics('train', train_metrics) if (not model_args.save_prefix_only): trainer.save_state() if (torch.cuda.is_available() and training_args.compute_memory): peak_memory = ((torch.cuda.max_memory_allocated() / (1024 ** 2)) / 1000) print('Memory utilization', peak_memory, 'GB') performance_metrics.update({'peak_memory': peak_memory}) if (training_args.compute_memory or training_args.compute_time): trainer.save_metrics('performance', performance_metrics) if ((model_args.shared_attn is True) and (model_args.ignore_target is False)): learned_embeddings = trainer.model.encoder.prefix_emb.clone().detach() results = {} if training_args.do_eval: logger.info('*** Evaluate ***') if (model_args.shared_attn is True): for (task, eval_dataset) in eval_datasets.items(): metrics = trainer.evaluate(eval_dataset=eval_dataset, max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, task=task) trainer.log_metrics(f'eval_{task}_', metrics) trainer.save_metrics(f"eval_{task.replace('/', ' ')}_", metrics) if training_args.wandb_log: wandb.log({f'eval_{task}_': metrics}) else: for (task, eval_dataset) in eval_datasets.items(): print('') print(task) print('') metrics = trainer.evaluate(eval_dataset=eval_dataset, max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, task=task) trainer.log_metrics(f'eval_{task}_', metrics) trainer.save_metrics(f"eval_{task.replace('/', ' ')}_", metrics) if training_args.wandb_log: wandb.log({f'eval_{task}_': metrics}) if model_args.save_prefix_only: checkpoints = glob.glob(os.path.join(training_args.output_dir, 'checkpoint-*')) for checkpoint_dir in checkpoints: if (not os.path.exists(os.path.join(checkpoint_dir, 'pytorch_model.bin'))): continue checkpoint_model = torch.load(os.path.join(os.path.join(checkpoint_dir, 'pytorch_model.bin'))) new_dir = '{}_prompt_only'.format(checkpoint_dir) os.mkdir(new_dir) for (name, param) in checkpoint_model.items(): if ((model_args.attn_prefix_tuning is False) and (('prefix_shared' in name) or ('prefix' in name))): shared_params = param torch.save(shared_params, os.path.join(training_args.output_dir, 'prefix_embeddings.pt')) elif ((model_args.attn_prefix_tuning is True) and (name == 'prefix_shared')): shared_params = param if (model_args.shared_attn is True): for i in range(config.num_target): torch.save(shared_params[i], os.path.join(new_dir, 'prefix_embeddings_{}.pt'.format(i))) else: torch.save(shared_params, os.path.join(new_dir, 'prefix_embeddings.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.attn_Wa.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(new_dir, 'attn_Wa_weights.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.attn_W_down.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(new_dir, 'attn_W_down.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.attn_W_up.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(new_dir, 'attn_W_up.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.layer_norm.weight' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(new_dir, 'layer_norm_weight.pt')) if ((model_args.attn_prefix_tuning is True) and ('encoder.layer_norm.bias' == name)): attn_weights_params = param torch.save(attn_weights_params, os.path.join(new_dir, 'layer_norm_bias.pt')) try: shutil.rmtree(checkpoint_dir) except OSError as e: print(('Error: %s : %s' % (checkpoint_dir, e.strerror))) if training_args.do_test: logger.info('*** Test ***') if (model_args.shared_attn is True): for (idx, (task, test_dataset)) in enumerate(test_datasets.items()): trainer.model.encoder.prefix_emb[0].data = learned_embeddings[idx] metrics = trainer.evaluate(eval_dataset=test_dataset, max_length=data_args.test_max_target_length, num_beams=data_args.num_beams, metric_key_prefix='test', task=task) trainer.log_metrics(f'test_{task}_', metrics) trainer.save_metrics(f"test_{task.replace('/', ' ')}_", metrics) else: for (task, test_dataset) in test_datasets.items(): metrics = trainer.evaluate(eval_dataset=test_dataset, max_length=data_args.test_max_target_length, num_beams=data_args.num_beams, metric_key_prefix='test', task=task) trainer.log_metrics(f'test_{task}_', metrics) trainer.save_metrics(f"test_{task.replace('/', ' ')}_", metrics) return results
class AtariLstmModel(torch.nn.Module): def __init__(self, image_shape, output_size, fc_sizes=512, lstm_size=512, use_maxpool=False, channels=None, kernel_sizes=None, strides=None, paddings=None): super().__init__() self.conv = Conv2dHeadModel(image_shape=image_shape, channels=(channels or [16, 32]), kernel_sizes=(kernel_sizes or [8, 4]), strides=(strides or [4, 2]), paddings=(paddings or [0, 1]), use_maxpool=use_maxpool, hidden_sizes=fc_sizes) self.lstm = torch.nn.LSTM(((self.conv.output_size + output_size) + 1), lstm_size) self.pi = torch.nn.Linear(lstm_size, output_size) self.value = torch.nn.Linear(lstm_size, 1) def forward(self, image, prev_action, prev_reward, init_rnn_state): img = image.type(torch.float) img = img.mul_((1.0 / 255)) (lead_dim, T, B, img_shape) = infer_leading_dims(img, 3) fc_out = self.conv(img.view((T * B), *img_shape)) lstm_input = torch.cat([fc_out.view(T, B, (- 1)), prev_action.view(T, B, (- 1)), prev_reward.view(T, B, 1)], dim=2) init_rnn_state = (None if (init_rnn_state is None) else tuple(init_rnn_state)) (lstm_out, (hn, cn)) = self.lstm(lstm_input, init_rnn_state) pi = F.softmax(self.pi(lstm_out.view((T * B), (- 1))), dim=(- 1)) v = self.value(lstm_out.view((T * B), (- 1))).squeeze((- 1)) (pi, v) = restore_leading_dims((pi, v), lead_dim, T, B) next_rnn_state = RnnState(h=hn, c=cn) return (pi, v, next_rnn_state)
class DiscLoss(): def name(self): return 'DiscLoss' def __init__(self, opt, tensor): self.criterionGAN = GANLoss(use_l1=False, tensor=tensor) self.fake_AB_pool = ImagePool(opt.pool_size) def get_g_loss(self, net, realA, fakeB): pred_fake = net.forward(fakeB) return self.criterionGAN(pred_fake, 1) def get_loss(self, net, realA, fakeB, realB): self.pred_fake = net.forward(fakeB.detach()) self.loss_D_fake = self.criterionGAN(self.pred_fake, 0) self.pred_real = net.forward(realB) self.loss_D_real = self.criterionGAN(self.pred_real, 1) self.loss_D = ((self.loss_D_fake + self.loss_D_real) * 0.5) return self.loss_D
def translate_and_store_swda_corpus_test_data(dataset, dataset_loading_function, dataset_file_path, translation_file_path, language, translate_whole_utterances=True): (talks_read, talk_names, _, _) = dataset_loading_function(dataset_file_path) if (dataset == 'MRDA'): test_set_idx = mrda_test_set_idx elif (dataset == 'SwDA'): test_set_idx = swda_test_set_idx else: print('Unknown dataset!') exit(0) if translate_whole_utterances: unit_str = 'u' else: unit_str = 'w' print(('# of talks read:' + str(len(talks_read)))) for i in range(len(talks_read)): if (talk_names[i] in test_set_idx): if translate_whole_utterances: (talk_read, talk_name) = translate_test_data_by_utterances([talks_read[i]], [talk_names[i]], test_set_idx, language) else: (talk_read, talk_name) = translate_test_data_by_words([talks_read[i]], [talk_names[i]], test_set_idx, language) talk_read = talk_read[0] talk_name = talk_name[0] print(('Storing file: %s' % talk_names[i])) fileName = ((((((translation_file_path + talk_name) + '_') + language) + '_') + unit_str) + '.txt') print(fileName) f = open(fileName, 'w') conversation = talk_read[0] f.write((str(len(conversation)) + '\n')) for utterance in conversation: f.write((str(len(utterance)) + '\n')) utterance_string = ' '.join(utterance) f.write((utterance_string + '\n')) tags = talks_read[i][1] for tag in tags: f.write((str(tag) + '\n')) f.close()
class PlotOut(Widget): def __init__(self, plots, plot_parameters, **kwargs): super().__init__(**kwargs) self.out = widgets.Output(layout={'border': '1px solid black', 'min_width': '300px', 'min_height': '300px', 'max_height': '600px', 'width': 'auto', 'height': 'auto', 'overflow': 'scroll'}) self.plot_selection_widget = DropdownSelectionWidget(plots, plots[0], 'Plots: ', plot_parameters) self.title = widgets.HTML(value='<H3>Plots</H3>') self.alpha_level = AlphaLevelWidget() def show(self): return widgets.VBox([self.title, self.out, self.plot_selection_widget.widget, self.alpha_level.show()]) def get_current_values(self): return (self.plot_selection_widget.get_value(), self.alpha_level.get_alpha_value(), self.plot_selection_widget.get_parameter_values()) def get_output(self): return self.out
class GPUFPSer(): def __init__(self, proj_name, args, pth_path): super(GPUFPSer, self).__init__() self.args = args self.to_pil = transforms.ToPILImage() self.proj_name = proj_name self.dev = torch.device('cuda:0') self.net = self.args[proj_name]['net']().to(self.dev) self.net.eval() self.test_image_transform = transforms.Compose([transforms.Resize((self.args['new_size'], self.args['new_size']), interpolation=Image.BILINEAR), transforms.ToTensor(), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]) if (pth_path != None): print(f'...{pth_path}') checkpoint = torch.load(pth_path) model_dict = self.net.state_dict() pretrained_dict = {k: v for (k, v) in checkpoint.items() if (k in model_dict)} model_dict.update(pretrained_dict) self.net.load_state_dict(model_dict) print('...') else: print('') def test(self, data_path, save_path): if save_path: print(f'{save_path}') check_mkdir(save_path) print(f'...{data_path}') img_path = os.path.join(data_path, 'Image') img_list = os.listdir(img_path) total_time = 0 tqdm_iter = tqdm(enumerate(img_list), total=len(img_list), leave=False) for (idx, img_name) in tqdm_iter: tqdm_iter.set_description(f'{self.proj_name}:te=>{(idx + 1)}') img_fullpath = os.path.join(img_path, img_name) test_image = Image.open(img_fullpath).convert('RGB') img_size = test_image.size test_image = self.test_image_transform(test_image) test_image = test_image.unsqueeze(0) test_image = test_image.to(self.dev) with torch.no_grad(): torch.cuda.synchronize() start_time = time.time() outputs = self.net(test_image) torch.cuda.synchronize() total_time += (time.time() - start_time) if save_path: outputs_np = outputs.squeeze(0).cpu().detach() out_img = self.to_pil(outputs_np).resize(img_size, Image.NEAREST) oimg_path = os.path.join(save_path, (img_name[:(- 4)] + '.png')) out_img.save(oimg_path) fps = (len(img_list) / total_time) return fps
def main_worker(gpu, args): args.gpu = gpu args.rank = gpu print(f'Process Launching at GPU {gpu}') if args.distributed: torch.cuda.set_device(args.gpu) dist.init_process_group(backend='nccl') if args.test_only: print(f'Building submit test loader at GPU {gpu}') split = f'submit_{gpu}' print('Loading', split) test_loader = get_loader(args, split=split, mode='val', batch_size=args.batch_size, distributed=False, gpu=args.gpu, workers=4, topk=args.valid_topk) train_loader = None val_loader = None trainer = Trainer(args, train_loader, val_loader, test_loader, train=False) dump_path = os.path.join(args.output, f'submit.json') trainer.predict(test_loader, dump_path=dump_path) if ('t5' in args.backbone): project_name = 'VLT5_GQA' elif ('bart' in args.backbone): project_name = 'VLBart_GQA' wandb.init(project=project_name) wandb.run.name = args.run_name wandb.config.update(args) src_dir = Path(__file__).resolve().parent base_path = str(src_dir.parent) src_dir = str(src_dir) wandb.save(os.path.join((src_dir + '/*.py')), base_path=base_path) wandb.save(dump_path, base_path=args.output) print('Uploaded', dump_path) else: print(f'Building train loader at GPU {gpu}') train_loader = get_loader(args, split=args.train, mode='train', batch_size=args.batch_size, distributed=args.distributed, gpu=args.gpu, workers=args.num_workers, topk=args.train_topk) if (args.valid_batch_size is not None): valid_batch_size = args.valid_batch_size else: valid_batch_size = args.batch_size print(f'Building val loader at GPU {gpu}') val_loader = get_loader(args, split=args.valid, mode='val', batch_size=valid_batch_size, distributed=args.distributed, gpu=args.gpu, workers=4, topk=args.valid_topk) print(f'Building test loader at GPU {gpu}') test_loader = get_loader(args, split=args.test, mode='val', batch_size=valid_batch_size, distributed=args.distributed, gpu=args.gpu, workers=4, topk=args.valid_topk) trainer = Trainer(args, train_loader, val_loader, test_loader, train=True) trainer.train()
def pre_process(entity, batchsize): processed_entity = [] for j in range(3): list = [] for i in range(batchsize): b = entity[i][j] list.append(b) c = torch.Tensor(list) processed_entity.append(c) return processed_entity
class ContinuousSamplerTaskPolicies(AbstractMctsPolicies): def __init__(self, task, Z={}, score_c=1.0, pw_C=1.0, pw_alpha=0.5, unordered=True, *args, **kwargs): self.task = task self.Z = Z sample = ContinuousTaskSample(task, Z, unordered) super(ContinuousSamplerTaskPolicies, self).__init__(*args, score=PriorProbabilityScore(score_c), widen=ProgressiveWiden(pw_C, pw_alpha), extract=MostVisitedExtract(), sample=sample, initialize=None, **kwargs)
class CalabresiDataset(Dataset): def __init__(self, csv_path, crop_type=None, crop_size=(192, 192), resize=None, eps: float=0.0001): super().__init__() self.csv_path = csv_path csv = pd.read_csv(csv_path) csv.rename(columns={'relapse_last30days': 'relapse'}, inplace=True) csv['relapse'] = csv['relapse'].map({np.nan: (- 1.0), 'N': 0.0, 'Y': 1.0}) csv['treatment_propagated'] = csv['treatment_propagated'].map({np.nan: (- 1.0), 'N': 0.0, 'Y': 1.0}) csv['treatment'] = csv['treatment'].map({np.nan: 0.0, 'none': 0.0, 'glatiramer acetate': 1.0, 'interferon beta': 2.0, 'natalizumab': 3.0, 'other': 4.0}) csv['duration'] = (csv['duration'].fillna(0.0) + eps) csv['edss'] = (csv['edss'].fillna(0.0) + eps) csv['msss'] = (csv['msss'].fillna(0.0) + eps) csv['fss'] = (csv['fss'].fillna(0.0) + eps) n_exist = (((~ csv['fss'].isnull()).astype(int) + (~ csv['msss'].isnull()).astype(int)) + (~ csv['edss'].isnull()).astype(int)) n_exist.replace(0.0, 1.0, inplace=True) csv['score'] = (((csv['edss'] + csv['msss']) + csv['fss']) / n_exist) csv['sex'] = csv['sex'].map({'M': 0.0, 'F': 1.0}) csv['type'] = csv['type'].map({'HC': 0.0, 'RRMS': 1.0, 'SPMS': 1.0, 'PPMS': 1.0}) csv['slice_ventricle_volume'] = (csv['slice_ventricle_volume'].astype(np.float32) + eps) csv['ventricle_volume'] = (csv['ventricle_volume'].astype(np.float32) + eps) csv['slice_brain_volume'] = (csv['slice_brain_volume'].astype(np.float32) + eps) csv['brain_volume'] = (csv['brain_volume'].astype(np.float32) + eps) csv['slice_lesion_volume'] = (csv['slice_lesion_volume'].astype(np.float32).fillna(0.0) + eps) csv['lesion_volume'] = (csv['lesion_volume'].astype(np.float32).fillna(0.0) + eps) csv['slice_number'] = csv['slice_number'].astype(np.float32) if csv.isnull().values.any(): raise ValueError(f'There is either an empty space, nan, or otherwise something wrong in the csv {csv_path}') self.csv = csv self.crop_type = crop_type self.crop_size = crop_size self.resize = resize def __len__(self): return len(self.csv) def __getitem__(self, index): item = self.csv.loc[index] item = item.to_dict() item = self._prepare_item(item) img_path = item['filename'] img = Image.open(img_path) transform_list = [] if (self.crop_type is not None): if (self.crop_type == 'center'): transform_list += [tv.transforms.CenterCrop(self.crop_size)] elif (self.crop_type == 'random'): transform_list += [tv.transforms.RandomCrop(self.crop_size)] else: raise ValueError(f'unknown crop type: {self.crop_type}') if self.resize: transform_list += [tv.transforms.Resize(self.resize)] transform_list += [tv.transforms.ToTensor()] img = tv.transforms.Compose(transform_list)(img) item['image'] = img return item def _prepare_item(item): item['age'] = torch.as_tensor(item['age'], dtype=torch.float32) item['sex'] = torch.as_tensor(item['type'], dtype=torch.float32) item['type'] = torch.as_tensor(item['type'], dtype=torch.float32) item['relapse'] = torch.as_tensor(item['relapse'], dtype=torch.float32) item['duration'] = torch.as_tensor(item['duration'], dtype=torch.float32) item['slice_brain_volume'] = torch.as_tensor(item['slice_brain_volume'], dtype=torch.float32) item['slice_ventricle_volume'] = torch.as_tensor(item['slice_ventricle_volume'], dtype=torch.float32) item['slice_lesion_volume'] = torch.as_tensor(item['slice_lesion_volume'], dtype=torch.float32) item['brain_volume'] = torch.as_tensor(item['brain_volume'], dtype=torch.float32) item['ventricle_volume'] = torch.as_tensor(item['ventricle_volume'], dtype=torch.float32) item['lesion_volume'] = torch.as_tensor(item['lesion_volume'], dtype=torch.float32) item['score'] = torch.as_tensor(item['score'], dtype=torch.float32) item['edss'] = torch.as_tensor(item['edss'], dtype=torch.float32) item['fss'] = torch.as_tensor(item['fss'], dtype=torch.float32) item['msss'] = torch.as_tensor(item['msss'], dtype=torch.float32) item['treatment'] = torch.as_tensor(item['treatment'], dtype=torch.float32) item['treatment_propagated'] = torch.as_tensor(item['treatment_propagated'], dtype=torch.float32) item['slice_number'] = torch.as_tensor(item['slice_number'], dtype=torch.float32) return item
def _dense_kernel_initializer(shape, dtype=None): (fan_in, fan_out) = _compute_fans(shape) stddev = (1.0 / np.sqrt(fan_in)) return K.random_uniform(shape, (- stddev), stddev, dtype)
class AllInOneVisualizer(): def __init__(self, evaluation: bool, model_names: [str], ns_prefix: str, visualize_every_x_iterations: int=1, visualize_crash: bool=False): self._evaluation = evaluation self.agent_visualization = rospy.Publisher(f'{ns_prefix}all_in_one_action_vis', Marker, queue_size=1) self.agent_visualization_trajectory = rospy.Publisher(f'{ns_prefix}all_in_one_action_trajectory_vis', Marker, queue_size=1) self.collision_visualization = rospy.Publisher(f'{ns_prefix}collision_vis', Marker, queue_size=1) self._setup_trajectory_marker() self._collisions_markers = [] self._model_names = model_names self.colors = [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0], [1.0, 1.0, 0.0], [1.0, 0.0, 1.0], [0.0, 1.0, 1.0]] self._collisions = 0 self._visualize_every_x_iterations = visualize_every_x_iterations self._current_iteration = 0 self._visualize_crash = visualize_crash def visualize_step(self, action: int, is_in_crash: bool, robot_pose: Pose2D): if ((self._current_iteration % self._visualize_every_x_iterations) == 0): self._visualize_action(action) self._visualize_episode_actions_as_path(action, robot_pose) if (self._visualize_crash and is_in_crash): self._visualize_collision(robot_pose) self._collisions += 1 self._current_iteration += 1 def reset_visualizer(self): if self._evaluation: self._action_trajectory_marker.points.clear() self._action_trajectory_marker.colors.clear() self.agent_visualization_trajectory.publish(self._action_trajectory_marker) self._remove_collisions_markers() self._collisions = 0 self._current_iteration = 0 def _setup_trajectory_marker(self): self._action_trajectory_marker = Marker() self._action_trajectory_marker.header.frame_id = 'map' self._action_trajectory_marker.id = 1001 self._action_trajectory_marker.action = visualization_msgs.msg.Marker.ADD self._action_trajectory_marker.type = visualization_msgs.msg.Marker.LINE_STRIP self._action_trajectory_marker.scale.z = 0 self._action_trajectory_marker.scale.x = 0.1 self._action_trajectory_marker.scale.y = 0 def _remove_collisions_markers(self): for m in self._collisions_markers: m.action = visualization_msgs.msg.Marker.DELETEALL self.collision_visualization.publish(m) self._collisions_markers = [] def _visualize_collision(self, robot_pose: Pose2D): marker = Marker() marker.header.frame_id = 'map' marker.header.stamp = rospy.get_rostime() marker.id = (1002 + self._collisions) marker.type = visualization_msgs.msg.Marker.TEXT_VIEW_FACING marker.action = visualization_msgs.msg.Marker.ADD marker.color.r = 1 marker.color.g = 1 marker.color.b = 0 marker.color.a = 1 marker.scale.z = 0.7 marker.pose.position.x = robot_pose.x marker.pose.position.y = robot_pose.y marker.text = 'Collision!' self._collisions_markers.append(marker) self.collision_visualization.publish(marker) def _visualize_episode_actions_as_path(self, action: int, robot_pose: Pose2D): marker = self._action_trajectory_marker marker.header.stamp = rospy.get_rostime() next_point = geometry_msgs.msg.Point() next_point.x = robot_pose.x next_point.y = robot_pose.y next_point.z = 0 marker.points.append(next_point) next_color = std_msgs.msg.ColorRGBA(self.colors[action][0], self.colors[action][1], self.colors[action][2], 1) marker.colors.append(next_color) self._action_trajectory_marker = marker self.agent_visualization_trajectory.publish(self._action_trajectory_marker) def _visualize_action(self, action: int): marker = Marker() marker.header.stamp = rospy.get_rostime() marker.header.frame_id = 'map' marker.header.stamp = rospy.get_rostime() marker.id = 1000 marker.type = visualization_msgs.msg.Marker.TEXT_VIEW_FACING marker.action = visualization_msgs.msg.Marker.ADD marker.color.r = self.colors[action][0] marker.color.g = self.colors[action][1] marker.color.b = self.colors[action][2] marker.color.a = 1 marker.pose.position.x = (- 0.5) marker.pose.position.y = (- 1.5) marker.scale.z = 1 marker.text = self._model_names[action] self.agent_visualization.publish(marker)
class Metric(metaclass=ABCMeta): _initialized = False def __init__(self, threaded=False, use_deque=False) -> None: super().__init__() self._initialized = True self._threaded = threaded if self._threaded: logger.trace(f'{self.__class__.__name__} spawn a thread') self._queue = ThreadQueue(use_deque=use_deque) self._worker = Thread(target=self._worker_func, name='threaded_worker', args=(self._queue,)) self._worker.start() atexit.register(self.close) def reset(self): pass def add(self, *args, **kwargs): assert self._initialized, f'{self.__class__.__name__} must be initialized by overriding __init__' if (not self._threaded): return self._add(*args, **kwargs) return self._add_queue(*args, **kwargs) def _add(self, *args, **kwargs): pass def _add_queue(self, *args, **kwargs): self._queue.put((args, kwargs)) def summary(self): return self._summary() def _summary(self): pass def _worker_func(self, input_queue: ThreadQueue): while True: try: (args, kwags) = input_queue.get() except IndexError: continue if isinstance(args[0], _StopToken): break self._add(*args, **kwags) def join(self): if (not self._threaded): return self.close() logger.trace(f'{self.__class__.__name__} join the thread') self._worker.join() logger.trace(f'{self.__class__.__name__} end the thread') def close(self): self._add_queue(_StopToken())
def sample_and_group(npoint, radius, nsample, xyz, points, new_xyz=None, knn=False, use_xyz=True): sample_idx = farthest_point_sample(npoint, xyz) if (new_xyz is None): new_xyz = gather_point(xyz, sample_idx) if knn: (_, idx) = knn_point(nsample, xyz, new_xyz) else: (idx, pts_cnt) = query_ball_point(radius, nsample, xyz, new_xyz) grouped_xyz = group_point(xyz, idx) grouped_xyz -= tf.tile(tf.expand_dims(new_xyz, 2), [1, 1, nsample, 1]) if (points is not None): grouped_points = group_point(points, idx) if use_xyz: new_points = tf.concat([grouped_xyz, grouped_points], axis=(- 1)) else: new_points = grouped_points else: new_points = grouped_xyz return (new_xyz, new_points, idx, sample_idx, grouped_xyz)
def train_model(model, criterion, optimizer, scheduler, cfg): best_acc = 0.0 best_map = 0.0 best_model_wts = copy.deepcopy(model.state_dict()) for epoch in range(1, cfg['max_epoch']): print(('-' * 60)) print('Epoch: {} / {}'.format(epoch, cfg['max_epoch'])) print(('-' * 60)) for phrase in ['train', 'test']: if (phrase == 'train'): model.train() else: model.eval() running_loss = 0.0 running_corrects = 0 (ft_all, lbl_all) = (None, None) for (i, (centers, corners, normals, neighbor_index, targets)) in enumerate(data_loader[phrase]): centers = centers.cuda() corners = corners.cuda() normals = normals.cuda() neighbor_index = neighbor_index.cuda() targets = targets.cuda() with torch.set_grad_enabled((phrase == 'train')): (outputs, feas) = model(centers, corners, normals, neighbor_index) (_, preds) = torch.max(outputs, 1) loss = criterion(outputs, targets) if (phrase == 'train'): optimizer.zero_grad() loss.backward() optimizer.step() if ((phrase == 'test') and cfg['retrieval_on']): ft_all = append_feature(ft_all, feas.detach().cpu()) lbl_all = append_feature(lbl_all, targets.detach().cpu(), flaten=True) running_loss += (loss.item() * centers.size(0)) running_corrects += torch.sum((preds == targets.data)) epoch_loss = (running_loss / len(data_set[phrase])) epoch_acc = (running_corrects.double() / len(data_set[phrase])) if (phrase == 'train'): print('{} Loss: {:.4f} Acc: {:.4f}'.format(phrase, epoch_loss, epoch_acc)) scheduler.step() if (phrase == 'test'): if (epoch_acc > best_acc): best_acc = epoch_acc best_model_wts = copy.deepcopy(model.state_dict()) print_info = '{} Loss: {:.4f} Acc: {:.4f} (best {:.4f})'.format(phrase, epoch_loss, epoch_acc, best_acc) if cfg['retrieval_on']: epoch_map = calculate_map(ft_all, lbl_all) if (epoch_map > best_map): best_map = epoch_map print_info += ' mAP: {:.4f}'.format(epoch_map) if ((epoch % cfg['save_steps']) == 0): torch.save(copy.deepcopy(model.state_dict()), os.path.join(cfg['ckpt_root'], '{}.pkl'.format(epoch))) print(print_info) print('Best val acc: {:.4f}'.format(best_acc)) print('Config: {}'.format(cfg)) return best_model_wts
def write_next_bytes(fid, data, format_char_sequence, endian_character='<'): if isinstance(data, (list, tuple)): bytes = struct.pack((endian_character + format_char_sequence), *data) else: bytes = struct.pack((endian_character + format_char_sequence), data) fid.write(bytes)
def train(args, train_dataset, model, tokenizer, train_highway=False): if (args.local_rank in [(- 1), 0]): tb_writer = SummaryWriter() args.train_batch_size = (args.per_gpu_train_batch_size * max(1, args.n_gpu)) train_sampler = (RandomSampler(train_dataset) if (args.local_rank == (- 1)) else DistributedSampler(train_dataset)) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size) if (args.max_steps > 0): t_total = args.max_steps args.num_train_epochs = ((args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps)) + 1) else: t_total = ((len(train_dataloader) // args.gradient_accumulation_steps) * args.num_train_epochs) no_decay = ['bias', 'LayerNorm.weight'] if train_highway: optimizer_grouped_parameters = [{'params': [p for (n, p) in model.named_parameters() if (('highway' in n) and (not any(((nd in n) for nd in no_decay))))], 'weight_decay': args.weight_decay}, {'params': [p for (n, p) in model.named_parameters() if (('highway' in n) and any(((nd in n) for nd in no_decay)))], 'weight_decay': 0.0}] else: optimizer_grouped_parameters = [{'params': [p for (n, p) in model.named_parameters() if (('highway' not in n) and (not any(((nd in n) for nd in no_decay))))], 'weight_decay': args.weight_decay}, {'params': [p for (n, p) in model.named_parameters() if (('highway' not in n) and any(((nd in n) for nd in no_decay)))], 'weight_decay': 0.0}] optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total) if args.fp16: try: from apex import amp except ImportError: raise ImportError('Please install apex from to use fp16 training.') (model, optimizer) = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level) if (args.n_gpu > 1): model = nn.DataParallel(model) if (args.local_rank != (- 1)): model = nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True) logger.info('***** Running training *****') logger.info(' Num examples = %d', len(train_dataset)) logger.info(' Num Epochs = %d', args.num_train_epochs) logger.info(' Instantaneous batch size per GPU = %d', args.per_gpu_train_batch_size) logger.info(' Total train batch size (w. parallel, distributed & accumulation) = %d', ((args.train_batch_size * args.gradient_accumulation_steps) * (torch.distributed.get_world_size() if (args.local_rank != (- 1)) else 1))) logger.info(' Gradient Accumulation steps = %d', args.gradient_accumulation_steps) logger.info(' Total optimization steps = %d', t_total) global_step = 0 (tr_loss, logging_loss) = (0.0, 0.0) model.zero_grad() train_iterator = trange(int(args.num_train_epochs), desc='Epoch', disable=(args.local_rank not in [(- 1), 0])) set_seed(args) for _ in train_iterator: epoch_iterator = tqdm(train_dataloader, desc='Iteration', disable=(args.local_rank not in [(- 1), 0])) for (step, batch) in enumerate(epoch_iterator): model.train() batch = tuple((t.to(args.device) for t in batch)) inputs = {'input_ids': batch[0], 'attention_mask': batch[1], 'labels': batch[3]} if (args.model_type != 'distilbert'): inputs['token_type_ids'] = (batch[2] if (args.model_type in ['bert', 'xlnet']) else None) inputs['train_highway'] = train_highway outputs = model(**inputs) loss = outputs[0] if (args.n_gpu > 1): loss = loss.mean() if (args.gradient_accumulation_steps > 1): loss = (loss / args.gradient_accumulation_steps) if args.fp16: with amp.scale_loss(loss, optimizer) as scaled_loss: scaled_loss.backward() else: loss.backward() tr_loss += loss.item() if (((step + 1) % args.gradient_accumulation_steps) == 0): if args.fp16: nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm) else: nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm) optimizer.step() scheduler.step() model.zero_grad() global_step += 1 if ((args.local_rank in [(- 1), 0]) and (args.logging_steps > 0) and ((global_step % args.logging_steps) == 0)): if ((args.local_rank == (- 1)) and args.evaluate_during_training): results = evaluate(args, model, tokenizer) for (key, value) in results.items(): tb_writer.add_scalar('eval_{}'.format(key), value, global_step) tb_writer.add_scalar('lr', scheduler.get_lr()[0], global_step) tb_writer.add_scalar('loss', ((tr_loss - logging_loss) / args.logging_steps), global_step) logging_loss = tr_loss if ((args.local_rank in [(- 1), 0]) and (args.save_steps > 0) and ((global_step % args.save_steps) == 0)): output_dir = os.path.join(args.output_dir, 'checkpoint-{}'.format(global_step)) if (not os.path.exists(output_dir)): os.makedirs(output_dir) model_to_save = (model.module if hasattr(model, 'module') else model) model_to_save.save_pretrained(output_dir) torch.save(args, os.path.join(output_dir, 'training_args.bin')) logger.info('Saving model checkpoint to %s', output_dir) if ((args.max_steps > 0) and (global_step > args.max_steps)): epoch_iterator.close() break if ((args.max_steps > 0) and (global_step > args.max_steps)): train_iterator.close() break if (args.local_rank in [(- 1), 0]): tb_writer.close() return (global_step, (tr_loss / global_step))
class JETS(AbsGANTTS): def __init__(self, idim: int, odim: int, sampling_rate: int=22050, generator_type: str='jets_generator', generator_params: Dict[(str, Any)]={'adim': 256, 'aheads': 2, 'elayers': 4, 'eunits': 1024, 'dlayers': 4, 'dunits': 1024, 'positionwise_layer_type': 'conv1d', 'positionwise_conv_kernel_size': 1, 'use_scaled_pos_enc': True, 'use_batch_norm': True, 'encoder_normalize_before': True, 'decoder_normalize_before': True, 'encoder_concat_after': False, 'decoder_concat_after': False, 'reduction_factor': 1, 'encoder_type': 'transformer', 'decoder_type': 'transformer', 'transformer_enc_dropout_rate': 0.1, 'transformer_enc_positional_dropout_rate': 0.1, 'transformer_enc_attn_dropout_rate': 0.1, 'transformer_dec_dropout_rate': 0.1, 'transformer_dec_positional_dropout_rate': 0.1, 'transformer_dec_attn_dropout_rate': 0.1, 'conformer_rel_pos_type': 'latest', 'conformer_pos_enc_layer_type': 'rel_pos', 'conformer_self_attn_layer_type': 'rel_selfattn', 'conformer_activation_type': 'swish', 'use_macaron_style_in_conformer': True, 'use_cnn_in_conformer': True, 'zero_triu': False, 'conformer_enc_kernel_size': 7, 'conformer_dec_kernel_size': 31, 'duration_predictor_layers': 2, 'duration_predictor_chans': 384, 'duration_predictor_kernel_size': 3, 'duration_predictor_dropout_rate': 0.1, 'energy_predictor_layers': 2, 'energy_predictor_chans': 384, 'energy_predictor_kernel_size': 3, 'energy_predictor_dropout': 0.5, 'energy_embed_kernel_size': 1, 'energy_embed_dropout': 0.5, 'stop_gradient_from_energy_predictor': False, 'pitch_predictor_layers': 5, 'pitch_predictor_chans': 384, 'pitch_predictor_kernel_size': 5, 'pitch_predictor_dropout': 0.5, 'pitch_embed_kernel_size': 1, 'pitch_embed_dropout': 0.5, 'stop_gradient_from_pitch_predictor': True, 'generator_out_channels': 1, 'generator_channels': 512, 'generator_global_channels': (- 1), 'generator_kernel_size': 7, 'generator_upsample_scales': [8, 8, 2, 2], 'generator_upsample_kernel_sizes': [16, 16, 4, 4], 'generator_resblock_kernel_sizes': [3, 7, 11], 'generator_resblock_dilations': [[1, 3, 5], [1, 3, 5], [1, 3, 5]], 'generator_use_additional_convs': True, 'generator_bias': True, 'generator_nonlinear_activation': 'LeakyReLU', 'generator_nonlinear_activation_params': {'negative_slope': 0.1}, 'generator_use_weight_norm': True, 'segment_size': 64, 'spks': (- 1), 'langs': (- 1), 'spk_embed_dim': None, 'spk_embed_integration_type': 'add', 'use_gst': False, 'gst_tokens': 10, 'gst_heads': 4, 'gst_conv_layers': 6, 'gst_conv_chans_list': [32, 32, 64, 64, 128, 128], 'gst_conv_kernel_size': 3, 'gst_conv_stride': 2, 'gst_gru_layers': 1, 'gst_gru_units': 128, 'init_type': 'xavier_uniform', 'init_enc_alpha': 1.0, 'init_dec_alpha': 1.0, 'use_masking': False, 'use_weighted_masking': False}, discriminator_type: str='hifigan_multi_scale_multi_period_discriminator', discriminator_params: Dict[(str, Any)]={'scales': 1, 'scale_downsample_pooling': 'AvgPool1d', 'scale_downsample_pooling_params': {'kernel_size': 4, 'stride': 2, 'padding': 2}, 'scale_discriminator_params': {'in_channels': 1, 'out_channels': 1, 'kernel_sizes': [15, 41, 5, 3], 'channels': 128, 'max_downsample_channels': 1024, 'max_groups': 16, 'bias': True, 'downsample_scales': [2, 2, 4, 4, 1], 'nonlinear_activation': 'LeakyReLU', 'nonlinear_activation_params': {'negative_slope': 0.1}, 'use_weight_norm': True, 'use_spectral_norm': False}, 'follow_official_norm': False, 'periods': [2, 3, 5, 7, 11], 'period_discriminator_params': {'in_channels': 1, 'out_channels': 1, 'kernel_sizes': [5, 3], 'channels': 32, 'downsample_scales': [3, 3, 3, 3, 1], 'max_downsample_channels': 1024, 'bias': True, 'nonlinear_activation': 'LeakyReLU', 'nonlinear_activation_params': {'negative_slope': 0.1}, 'use_weight_norm': True, 'use_spectral_norm': False}}, generator_adv_loss_params: Dict[(str, Any)]={'average_by_discriminators': False, 'loss_type': 'mse'}, discriminator_adv_loss_params: Dict[(str, Any)]={'average_by_discriminators': False, 'loss_type': 'mse'}, feat_match_loss_params: Dict[(str, Any)]={'average_by_discriminators': False, 'average_by_layers': False, 'include_final_outputs': True}, mel_loss_params: Dict[(str, Any)]={'fs': 22050, 'n_fft': 1024, 'hop_length': 256, 'win_length': None, 'window': 'hann', 'n_mels': 80, 'fmin': 0, 'fmax': None, 'log_base': None}, lambda_adv: float=1.0, lambda_mel: float=45.0, lambda_feat_match: float=2.0, lambda_var: float=1.0, lambda_align: float=2.0, cache_generator_outputs: bool=True): assert check_argument_types() super().__init__() generator_class = AVAILABLE_GENERATERS[generator_type] generator_params.update(idim=idim, odim=odim) self.generator = generator_class(**generator_params) discriminator_class = AVAILABLE_DISCRIMINATORS[discriminator_type] self.discriminator = discriminator_class(**discriminator_params) self.generator_adv_loss = GeneratorAdversarialLoss(**generator_adv_loss_params) self.discriminator_adv_loss = DiscriminatorAdversarialLoss(**discriminator_adv_loss_params) self.feat_match_loss = FeatureMatchLoss(**feat_match_loss_params) self.mel_loss = MelSpectrogramLoss(**mel_loss_params) self.var_loss = VarianceLoss() self.forwardsum_loss = ForwardSumLoss() self.lambda_adv = lambda_adv self.lambda_mel = lambda_mel self.lambda_feat_match = lambda_feat_match self.lambda_var = lambda_var self.lambda_align = lambda_align self.cache_generator_outputs = cache_generator_outputs self._cache = None self.fs = sampling_rate self.spks = self.generator.spks self.langs = self.generator.langs self.spk_embed_dim = self.generator.spk_embed_dim def require_raw_speech(self): return True def require_vocoder(self): return False def forward(self, text: torch.Tensor, text_lengths: torch.Tensor, feats: torch.Tensor, feats_lengths: torch.Tensor, speech: torch.Tensor, speech_lengths: torch.Tensor, sids: Optional[torch.Tensor]=None, spembs: Optional[torch.Tensor]=None, lids: Optional[torch.Tensor]=None, forward_generator: bool=True, **kwargs) -> Dict[(str, Any)]: if forward_generator: return self._forward_generator(text=text, text_lengths=text_lengths, feats=feats, feats_lengths=feats_lengths, speech=speech, speech_lengths=speech_lengths, sids=sids, spembs=spembs, lids=lids, **kwargs) else: return self._forward_discrminator(text=text, text_lengths=text_lengths, feats=feats, feats_lengths=feats_lengths, speech=speech, speech_lengths=speech_lengths, sids=sids, spembs=spembs, lids=lids, **kwargs) def _forward_generator(self, text: torch.Tensor, text_lengths: torch.Tensor, feats: torch.Tensor, feats_lengths: torch.Tensor, speech: torch.Tensor, speech_lengths: torch.Tensor, sids: Optional[torch.Tensor]=None, spembs: Optional[torch.Tensor]=None, lids: Optional[torch.Tensor]=None, **kwargs) -> Dict[(str, Any)]: batch_size = text.size(0) speech = speech.unsqueeze(1) reuse_cache = True if ((not self.cache_generator_outputs) or (self._cache is None)): reuse_cache = False outs = self.generator(text=text, text_lengths=text_lengths, feats=feats, feats_lengths=feats_lengths, sids=sids, spembs=spembs, lids=lids, **kwargs) else: outs = self._cache if (self.training and self.cache_generator_outputs and (not reuse_cache)): self._cache = outs (speech_hat_, bin_loss, log_p_attn, start_idxs, d_outs, ds, p_outs, ps, e_outs, es) = outs speech_ = get_segments(x=speech, start_idxs=(start_idxs * self.generator.upsample_factor), segment_size=(self.generator.segment_size * self.generator.upsample_factor)) p_hat = self.discriminator(speech_hat_) with torch.no_grad(): p = self.discriminator(speech_) mel_loss = self.mel_loss(speech_hat_, speech_) adv_loss = self.generator_adv_loss(p_hat) feat_match_loss = self.feat_match_loss(p_hat, p) (dur_loss, pitch_loss, energy_loss) = self.var_loss(d_outs, ds, p_outs, ps, e_outs, es, text_lengths) forwardsum_loss = self.forwardsum_loss(log_p_attn, text_lengths, feats_lengths) mel_loss = (mel_loss * self.lambda_mel) adv_loss = (adv_loss * self.lambda_adv) feat_match_loss = (feat_match_loss * self.lambda_feat_match) g_loss = ((mel_loss + adv_loss) + feat_match_loss) var_loss = (((dur_loss + pitch_loss) + energy_loss) * self.lambda_var) align_loss = ((forwardsum_loss + bin_loss) * self.lambda_align) loss = ((g_loss + var_loss) + align_loss) stats = dict(generator_loss=loss.item(), generator_g_loss=g_loss.item(), generator_var_loss=var_loss.item(), generator_align_loss=align_loss.item(), generator_g_mel_loss=mel_loss.item(), generator_g_adv_loss=adv_loss.item(), generator_g_feat_match_loss=feat_match_loss.item(), generator_var_dur_loss=dur_loss.item(), generator_var_pitch_loss=pitch_loss.item(), generator_var_energy_loss=energy_loss.item(), generator_align_forwardsum_loss=forwardsum_loss.item(), generator_align_bin_loss=bin_loss.item()) (loss, stats, weight) = force_gatherable((loss, stats, batch_size), loss.device) if (reuse_cache or (not self.training)): self._cache = None return {'loss': loss, 'stats': stats, 'weight': weight, 'optim_idx': 0} def _forward_discrminator(self, text: torch.Tensor, text_lengths: torch.Tensor, feats: torch.Tensor, feats_lengths: torch.Tensor, speech: torch.Tensor, speech_lengths: torch.Tensor, sids: Optional[torch.Tensor]=None, spembs: Optional[torch.Tensor]=None, lids: Optional[torch.Tensor]=None, **kwargs) -> Dict[(str, Any)]: batch_size = text.size(0) speech = speech.unsqueeze(1) reuse_cache = True if ((not self.cache_generator_outputs) or (self._cache is None)): reuse_cache = False outs = self.generator(text=text, text_lengths=text_lengths, feats=feats, feats_lengths=feats_lengths, sids=sids, spembs=spembs, lids=lids, **kwargs) else: outs = self._cache if (self.cache_generator_outputs and (not reuse_cache)): self._cache = outs (speech_hat_, _, _, start_idxs, *_) = outs speech_ = get_segments(x=speech, start_idxs=(start_idxs * self.generator.upsample_factor), segment_size=(self.generator.segment_size * self.generator.upsample_factor)) p_hat = self.discriminator(speech_hat_.detach()) p = self.discriminator(speech_) (real_loss, fake_loss) = self.discriminator_adv_loss(p_hat, p) loss = (real_loss + fake_loss) stats = dict(discriminator_loss=loss.item(), discriminator_real_loss=real_loss.item(), discriminator_fake_loss=fake_loss.item()) (loss, stats, weight) = force_gatherable((loss, stats, batch_size), loss.device) if (reuse_cache or (not self.training)): self._cache = None return {'loss': loss, 'stats': stats, 'weight': weight, 'optim_idx': 1} def inference(self, text: torch.Tensor, feats: Optional[torch.Tensor]=None, pitch: Optional[torch.Tensor]=None, energy: Optional[torch.Tensor]=None, use_teacher_forcing: bool=False, **kwargs) -> Dict[(str, torch.Tensor)]: text = text[None] text_lengths = torch.tensor([text.size(1)], dtype=torch.long, device=text.device) if use_teacher_forcing: assert (feats is not None) feats = feats[None] feats_lengths = torch.tensor([feats.size(1)], dtype=torch.long, device=feats.device) pitch = pitch[None] energy = energy[None] (wav, dur) = self.generator.inference(text=text, text_lengths=text_lengths, feats=feats, feats_lengths=feats_lengths, pitch=pitch, energy=energy, use_teacher_forcing=use_teacher_forcing, **kwargs) else: (wav, dur) = self.generator.inference(text=text, text_lengths=text_lengths, **kwargs) return dict(wav=wav.view((- 1)), duration=dur[0])
class GradOutputMonitor(BaseMonitor): def __init__(self, net: nn.Module, instance: (Any or tuple)=None, function_on_grad_output: Callable=(lambda x: x)): super().__init__() self.function_on_grad_output = function_on_grad_output for (name, m) in net.named_modules(): if isinstance(m, instance): self.monitored_layers.append(name) self.name_records_index[name] = [] if (torch.__version__ >= torch.torch_version.TorchVersion('1.8.0')): self.hooks.append(m.register_full_backward_hook(self.create_hook(name))) else: self.hooks.append(m.register_backward_hook(self.create_hook(name))) def create_hook(self, name): def hook(m, grad_input, grad_output): if self.is_enable(): self.name_records_index[name].append(self.records.__len__()) self.records.append(self.function_on_grad_output(unpack_len1_tuple(grad_output))) return hook
class DataArguments(): dataset_name: Optional[str] = field(default=None, metadata={'help': 'The name of the dataset to use (via the datasets library).'}) dataset_config_name: Optional[str] = field(default=None, metadata={'help': 'The configuration name of the dataset to use (via the datasets library).'}) train_file: Optional[str] = field(default=None, metadata={'help': 'The input training data file (a text file).'}) validation_file: Optional[str] = field(default=None, metadata={'help': 'An optional input evaluation data file to evaluate the perplexity on (a text file).'}) max_seq_length: Optional[int] = field(default=512, metadata={'help': 'The maximum total input sequence length after tokenization. Sequences longer than this will be truncated.'}) validation_split_percentage: Optional[int] = field(default=0, metadata={'help': "The percentage of the train set used as validation set in case there's no validation split"}) overwrite_cache: bool = field(default=False, metadata={'help': 'Overwrite the cached preprocessed datasets or not.'}) pad_to_max_length: bool = field(default=False, metadata={'help': 'Whether to pad all samples to `max_seq_length`. If False, will pad the samples dynamically when batching to the maximum length in the batch.'}) max_train_samples: Optional[int] = field(default=None, metadata={'help': 'For debugging purposes or quicker training, truncate the number of training examples to this value if set.'}) max_eval_samples: Optional[int] = field(default=None, metadata={'help': 'For debugging purposes or quicker training, truncate the number of evaluation examples to this value if set.'}) keep_in_memory: bool = field(default=False, metadata={'help': 'Whether to keep in memory the loaded dataset. Defaults to False.'}) dataset_seed: int = field(default=42, metadata={'help': 'Seed to use in dataset processing, different seeds might yield different datasets. This seed and the seed in training arguments are not related'}) dataset_cache_directory: Optional[str] = field(default=None, metadata={'help': 'Path to directory where the processed dataset will be saved. If path exists, try to load processed dataset from this path.'}) dataset_concatenation: Optional[bool] = field(default=False, metadata={'help': 'Whether to concatenate the sentence for more efficient training.'}) special_tokens: Optional[List[str]] = field(default=None, metadata={'help': 'The list of special tokens to add in tokenizer.'}) max_source_length: Optional[int] = field(default=384, metadata={'help': 'The maximum total input sequence length after tokenization. Sequences longer than this will be truncated, sequences shorter will be padded.'}) max_target_length: Optional[int] = field(default=128, metadata={'help': 'The maximum total sequence length for target text after tokenization. Sequences longer than this will be truncated, sequences shorter will be padded.'}) num_beams: Optional[int] = field(default=4, metadata={'help': 'Number of beams to use for evaluation. This argument will be passed to ``model.generate``, which is used during ``evaluate`` and ``predict``.'}) streaming: bool = field(default=False, metadata={'help': 'Enable streaming mode'}) preprocessing_num_workers: Optional[int] = field(default=None, metadata={'help': 'The number of processes to use for the preprocessing.'}) def __post_init__(self): if self.streaming: require_version('datasets>=2.0.0', 'The streaming feature requires `datasets>=2.0.0`') if ((self.dataset_name is None) and (self.train_file is None) and (self.validation_file is None)): raise ValueError('Need either a dataset name or a training/validation file.') else: if (self.train_file is not None): extension = self.train_file.split('.')[(- 1)] assert (extension in ['csv', 'json', 'txt']), '`train_file` should be a csv, a json or a txt file.' if (self.validation_file is not None): extension = self.validation_file.split('.')[(- 1)] assert (extension in ['csv', 'json', 'txt']), '`validation_file` should be a csv, a json or a txt file.'
def parse_args(): parser = argparse.ArgumentParser(description='Process arguments') parser.add_argument('--model_type', type=str, required=True) parser.add_argument('--datasize', type=int, default=200, required=False) parser.add_argument('--epoch', type=int, default=2, required=False) parser.add_argument('--hiddien_size', type=int, default=32, required=False) parser.add_argument('--head_num', type=int, default=4, required=False) parser.add_argument('--layer_num', type=int, default=3, required=False) parser.add_argument('--seq_length', type=int, default=16, required=False) parser.add_argument('--batchsize', type=int, default=8, required=False) parser.add_argument('--distributed', default=False, action='store_true') parser.add_argument('--user_created_dataloader', default=False, action='store_true') parser.add_argument('--load_strategy', default=False, action='store_true') parser.add_argument('--optim_grouped_params', default=False, action='store_true') parser.add_argument('--log_interval', type=int, default=10, required=False) parser.add_argument('--use_fsdp', default=False, action='store_true') parser.add_argument('--use_amp', default=False, action='store_true') parser.add_argument('--use_checkpointing', default=False, action='store_true') parser.add_argument('--use_module_replace', default=False, action='store_true') return parser.parse_args()
def single_or_none(sequence: List[E]) -> Optional[E]: assert (len(sequence) <= 1) return next(iter(sequence), None)
def combine_collision_reports(r1: CollisionReport, r2: CollisionReport) -> CollisionReport: if (r1.players.keys() != r2.players.keys()): raise ZValueError('Cannot combine collision reports with different players', report1=r1, report2=r2) (first_report, second_report) = ((r1, r2) if (r1.at_time <= r2.at_time) else (r2, r1)) combined_players_report: dict[(PlayerName, CollisionReportPlayer)] = {} for p in first_report.players: (r1_player, r2_player) = (first_report.players[p], second_report.players[p]) combined_players_report[p] = replace(r1_player, locations=(r1_player.locations + r2_player.locations), velocity_after=r2_player.velocity_after, energy_delta=(r1_player.energy_delta + r2_player.energy_delta)) return replace(first_report, players=combined_players_report)
class PTBTokenizer(): def tokenize(self, captions_for_image): cmd = ['java', '-cp', STANFORD_CORENLP_3_4_1_JAR, 'edu.stanford.nlp.process.PTBTokenizer', '-preserveLines', '-lowerCase'] final_tokenized_captions_for_image = {} image_id = [k for (k, v) in captions_for_image.items() for _ in range(len(v))] sentences = '\n'.join([c['caption'].replace('\n', ' ') for (k, v) in captions_for_image.items() for c in v]) path_to_jar_dirname = os.path.dirname(os.path.abspath(__file__)) tmp_file_name = (((((path_to_jar_dirname + '/temp_file') + str(random.randint(0, 999999))) + '_pid_') + str(os.getpid())) + '.tmp') with open(tmp_file_name, 'w') as tmp_file: tmp_file.write(sentences) cmd.append(tmp_file_name) p_tokenizer = subprocess.Popen(cmd, cwd=path_to_jar_dirname, stdout=subprocess.PIPE, stderr=subprocess.DEVNULL) token_lines = p_tokenizer.communicate(input=sentences.rstrip())[0] token_lines = token_lines.decode('utf-8') lines = token_lines.split('\n') if os.path.isfile(tmp_file_name): os.remove(tmp_file_name) for (k, line) in zip(image_id, lines): if (not (k in final_tokenized_captions_for_image)): final_tokenized_captions_for_image[k] = [] tokenized_caption = ' '.join([w for w in line.rstrip().split(' ') if (w not in PUNCTUATIONS)]) final_tokenized_captions_for_image[k].append(tokenized_caption) return final_tokenized_captions_for_image
class LossHistory(): def __init__(self): self.steps = [] self.loss_train = [] self.loss_test = [] self.metrics_test = [] self.loss_weights = None def set_loss_weights(self, loss_weights): self.loss_weights = loss_weights def append(self, step, loss_train, loss_test, metrics_test): self.steps.append(step) self.loss_train.append(loss_train) if (loss_test is None): loss_test = self.loss_test[(- 1)] if (metrics_test is None): metrics_test = self.metrics_test[(- 1)] self.loss_test.append(loss_test) self.metrics_test.append(metrics_test)
class RepetitionPenaltyLogitsProcessor(metaclass=DummyObject): _backends = ['torch'] def __init__(self, *args, **kwargs): requires_backends(self, ['torch'])
def load_sem_seg(gt_root, image_root, gt_ext='png', image_ext='jpg'): def file2id(folder_path, file_path): image_id = os.path.normpath(os.path.relpath(file_path, start=folder_path)) image_id = os.path.splitext(image_id)[0] return image_id input_files = sorted((os.path.join(image_root, f) for f in PathManager.ls(image_root) if f.endswith(image_ext)), key=(lambda file_path: file2id(image_root, file_path))) gt_files = sorted((os.path.join(gt_root, f) for f in PathManager.ls(gt_root) if f.endswith(gt_ext)), key=(lambda file_path: file2id(gt_root, file_path))) assert (len(gt_files) > 0), 'No annotations found in {}.'.format(gt_root) if (len(input_files) != len(gt_files)): logger.warn('Directory {} and {} has {} and {} files, respectively.'.format(image_root, gt_root, len(input_files), len(gt_files))) input_basenames = [os.path.basename(f)[:(- len(image_ext))] for f in input_files] gt_basenames = [os.path.basename(f)[:(- len(gt_ext))] for f in gt_files] intersect = list((set(input_basenames) & set(gt_basenames))) intersect = sorted(intersect) logger.warn('Will use their intersection of {} files.'.format(len(intersect))) input_files = [os.path.join(image_root, (f + image_ext)) for f in intersect] gt_files = [os.path.join(gt_root, (f + gt_ext)) for f in intersect] logger.info('Loaded {} images with semantic segmentation from {}'.format(len(input_files), image_root)) dataset_dicts = [] for (img_path, gt_path) in zip(input_files, gt_files): local_path = PathManager.get_local_path(gt_path) (w, h) = imagesize.get(local_path) record = {} record['file_name'] = img_path record['sem_seg_file_name'] = gt_path record['height'] = h record['width'] = w dataset_dicts.append(record) return dataset_dicts
def _two_particle_antisymmetric_spatial_wavefn(unused_params, x): del unused_params wavefn_amp = (x[(..., 0, 0)] - x[(..., 1, 0)]) return array_to_slog(wavefn_amp)
class CustomDataset(Dataset): def __init__(self, size=1000): (x, y) = get_x_y(size=size) self.x = torch.from_numpy(x).float() self.y = torch.from_numpy(y).float() def __len__(self): return self.x.shape[0] def __getitem__(self, idx): return (self.x[idx], self.y[idx])
def get_network(weights): if (weights in WEIGHTS_URLS.keys()): arch_params = WEIGHTS_URLS[weights]['arch_params'] url = WEIGHTS_URLS[weights]['url'] name = WEIGHTS_URLS[weights]['name'] else: raise ValueError('Available RRDN network weights: {}'.format(list(WEIGHTS_URLS.keys()))) c_dim = 3 kernel_size = 3 return (arch_params, c_dim, kernel_size, url, name)
def encoder_init(m): if isinstance(m, torch.nn.Conv1d): torch.nn.init.xavier_uniform_(m.weight, torch.nn.init.calculate_gain('relu'))
def preprocess_function(examples): inputs = examples[text_column] targets = examples[label_column] model_inputs = tokenizer(inputs, max_length=max_length, padding='max_length', truncation=True, return_tensors='pt') labels = tokenizer(targets, max_length=2, padding='max_length', truncation=True, return_tensors='pt') labels = labels['input_ids'] labels[(labels == tokenizer.pad_token_id)] = (- 100) model_inputs['labels'] = labels return model_inputs
_sentencepiece _tokenizers class ReformerTokenizationTest(TokenizerTesterMixin, unittest.TestCase): tokenizer_class = ReformerTokenizer rust_tokenizer_class = ReformerTokenizerFast test_rust_tokenizer = True test_seq2seq = False test_sentencepiece = True def setUp(self): super().setUp() tokenizer = ReformerTokenizer(SAMPLE_VOCAB, keep_accents=True) tokenizer.save_pretrained(self.tmpdirname) def test_convert_token_and_id(self): token = '<s>' token_id = 1 self.assertEqual(self.get_tokenizer()._convert_token_to_id(token), token_id) self.assertEqual(self.get_tokenizer()._convert_id_to_token(token_id), token) def test_get_vocab(self): vocab_keys = list(self.get_tokenizer().get_vocab().keys()) self.assertEqual(vocab_keys[0], '<unk>') self.assertEqual(vocab_keys[1], '<s>') self.assertEqual(vocab_keys[(- 1)], 'j') self.assertEqual(len(vocab_keys), 1000) def test_vocab_size(self): self.assertEqual(self.get_tokenizer().vocab_size, 1000) def test_rust_and_python_full_tokenizers(self): if (not self.test_rust_tokenizer): return tokenizer = self.get_tokenizer() rust_tokenizer = self.get_rust_tokenizer() sequence = 'I was born in 92000, and this is false.' tokens = tokenizer.tokenize(sequence) rust_tokens = rust_tokenizer.tokenize(sequence) self.assertListEqual(tokens, rust_tokens) ids = tokenizer.encode(sequence, add_special_tokens=False) rust_ids = rust_tokenizer.encode(sequence, add_special_tokens=False) self.assertListEqual(ids, rust_ids) rust_tokenizer = self.get_rust_tokenizer() ids = tokenizer.encode(sequence) rust_ids = rust_tokenizer.encode(sequence) self.assertListEqual(ids, rust_ids) def test_padding(self, max_length=15): for (tokenizer, pretrained_name, kwargs) in self.tokenizers_list: with self.subTest(f'{tokenizer.__class__.__name__} ({pretrained_name})'): tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs) s = 'This is a simple input' s2 = ['This is a simple input 1', 'This is a simple input 2'] p = ('This is a simple input', 'This is a pair') p2 = [('This is a simple input 1', 'This is a simple input 2'), ('This is a simple pair 1', 'This is a simple pair 2')] self.assertRaises(ValueError, tokenizer_r.encode, s, max_length=max_length, padding='max_length') self.assertRaises(ValueError, tokenizer_r.encode_plus, s, max_length=max_length, padding='max_length') self.assertRaises(ValueError, tokenizer_r.batch_encode_plus, s2, max_length=max_length, padding='max_length') self.assertRaises(ValueError, tokenizer_r.encode, p, max_length=max_length, padding='max_length') self.assertRaises(ValueError, tokenizer_r.encode_plus, p, max_length=max_length, padding='max_length') self.assertRaises(ValueError, tokenizer_r.batch_encode_plus, p2, max_length=max_length, padding='max_length') def test_padding_different_model_input_name(self): pass def test_full_tokenizer(self): tokenizer = ReformerTokenizer(SAMPLE_VOCAB, keep_accents=True) tokens = tokenizer.tokenize('This is a test') self.assertListEqual(tokens, ['This', 'is', 'a', 't', 'est']) self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [285, 46, 10, 170, 382]) tokens = tokenizer.tokenize('I was born in 92000, and this is false.') self.assertListEqual(tokens, [(SPIECE_UNDERLINE + 'I'), (SPIECE_UNDERLINE + 'was'), (SPIECE_UNDERLINE + 'b'), 'or', 'n', (SPIECE_UNDERLINE + 'in'), (SPIECE_UNDERLINE + ''), '9', '2', '0', '0', '0', ',', (SPIECE_UNDERLINE + 'and'), (SPIECE_UNDERLINE + 'this'), (SPIECE_UNDERLINE + 'is'), (SPIECE_UNDERLINE + 'f'), 'al', 's', 'e', '.']) ids = tokenizer.convert_tokens_to_ids(tokens) self.assertListEqual(ids, [8, 21, 84, 55, 24, 19, 7, 0, 602, 347, 347, 347, 3, 12, 66, 46, 72, 80, 6, 0, 4]) back_tokens = tokenizer.convert_ids_to_tokens(ids) self.assertListEqual(back_tokens, [(SPIECE_UNDERLINE + 'I'), (SPIECE_UNDERLINE + 'was'), (SPIECE_UNDERLINE + 'b'), 'or', 'n', (SPIECE_UNDERLINE + 'in'), (SPIECE_UNDERLINE + ''), '<unk>', '2', '0', '0', '0', ',', (SPIECE_UNDERLINE + 'and'), (SPIECE_UNDERLINE + 'this'), (SPIECE_UNDERLINE + 'is'), (SPIECE_UNDERLINE + 'f'), 'al', 's', '<unk>', '.']) _property def big_tokenizer(self): return ReformerTokenizer.from_pretrained('google/reformer-crime-and-punishment') def test_tokenization_base_easy_symbols(self): symbols = 'Hello World!' original_tokenizer_encodings = [126, 32, 262, 152, 38, 72, 287] self.assertListEqual(original_tokenizer_encodings, self.big_tokenizer.encode(symbols)) def test_tokenization_base_hard_symbols(self): symbols = 'This is a very long text with a lot of weird characters, such as: . , ~ ? ( ) " [ ] ! : - . Also we will add words that should not exsist and be tokenized to <unk>, such as saoneuhaoesuth' original_tokenizer_encodings = [108, 265, 24, 111, 4, 258, 156, 35, 28, 275, 3, 259, 297, 260, 84, 4, 35, 110, 44, 8, 259, 91, 268, 21, 11, 209, 274, 109, 266, 277, 117, 86, 93, 315, 258, 278, 258, 277, 258, 0, 258, 288, 258, 319, 258, 0, 258, 0, 258, 0, 258, 0, 258, 287, 258, 315, 258, 289, 258, 278, 99, 269, 266, 262, 8, 259, 241, 4, 217, 230, 268, 266, 55, 168, 106, 75, 193, 266, 223, 27, 49, 26, 282, 25, 264, 299, 19, 26, 0, 258, 277, 117, 86, 93, 176, 183, 270, 11, 262, 42, 61, 265] self.assertListEqual(original_tokenizer_encodings, self.big_tokenizer.encode(symbols)) _torch def test_torch_encode_plus_sent_to_model(self): import torch from transformers import ReformerConfig, ReformerModel first_ten_tokens = list(self.big_tokenizer.get_vocab().keys())[:10] sequence = ' '.join(first_ten_tokens) encoded_sequence = self.big_tokenizer.encode_plus(sequence, return_tensors='pt') batch_encoded_sequence = self.big_tokenizer.batch_encode_plus([sequence, sequence], return_tensors='pt') config = ReformerConfig() config.axial_pos_shape = encoded_sequence['input_ids'].shape model = ReformerModel(config) assert (model.get_input_embeddings().weight.shape[0] >= self.big_tokenizer.vocab_size) with torch.no_grad(): model(**encoded_sequence) model(**batch_encoded_sequence) def test_tokenizer_integration(self): expected_encoding = {'input_ids': [[108, 265, 24, 111, 4, 258, 156, 7, 51, 279, 58, 7, 76, 25, 69, 278], [140, 243, 264, 134, 17, 267, 77, 263, 22, 262, 297, 258, 304, 177, 279, 266, 14, 89, 13, 35, 261, 299, 272, 137, 275, 278]], 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]} sequences = ['This is a very simple sentence.', 'The quick brown fox jumps over the lazy dog.'] self.tokenizer_integration_test_util(expected_encoding=expected_encoding, model_name='google/reformer-crime-and-punishment', revision='0e6c3decb8211d49bfdc8b0448b3f5a', padding=False, sequences=sequences)
def cal_acc(x, gt): x = x.squeeze(0) gt = gt.squeeze(0) if (x.shape[0] > x.shape[1]): x = x.t() gt = gt.t() b = (x.max(1).indices == gt.max(1).indices) if (b.shape[0] != x.shape[0]): raise RuntimeError('b.shape[0] != x.shape[0]') return (int(b.sum()), b.shape[0])
class BoundedArchive(Archive[S]): def __init__(self, maximum_size: int, comparator: Comparator[S]=None, density_estimator: DensityEstimator=None, dominance_comparator: Comparator[S]=DominanceComparator()): super(BoundedArchive, self).__init__() self.maximum_size = maximum_size self.comparator = comparator self.density_estimator = density_estimator self.non_dominated_solution_archive = NonDominatedSolutionsArchive(dominance_comparator=dominance_comparator) self.solution_list = self.non_dominated_solution_archive.solution_list def compute_density_estimator(self): self.density_estimator.compute_density_estimator(self.solution_list) def add(self, solution: S) -> bool: success = self.non_dominated_solution_archive.add(solution) if success: if (self.size() > self.maximum_size): self.compute_density_estimator() (worst_solution, index_to_remove) = self.__find_worst_solution(self.solution_list) self.solution_list.pop(index_to_remove) return success def __find_worst_solution(self, solution_list: List[S]) -> S: if (solution_list is None): raise Exception('The solution list is None') elif (len(solution_list) == 0): raise Exception('The solution list is empty') worst_solution = solution_list[0] index_to_remove = 0 for (solution_index, solution) in enumerate(solution_list[1:]): if (self.comparator.compare(worst_solution, solution) < 0): worst_solution = solution index_to_remove = (solution_index + 1) return (worst_solution, index_to_remove)
def set_cursor_enter_callback(window, cbfun): window_addr = ctypes.cast(ctypes.pointer(window), ctypes.POINTER(ctypes.c_long)).contents.value if (window_addr in _cursor_enter_callback_repository): previous_callback = _cursor_enter_callback_repository[window_addr] else: previous_callback = None if (cbfun is None): cbfun = 0 c_cbfun = _GLFWcursorenterfun(cbfun) _cursor_enter_callback_repository[window_addr] = (cbfun, c_cbfun) cbfun = c_cbfun _glfw.glfwSetCursorEnterCallback(window, cbfun) if ((previous_callback is not None) and (previous_callback[0] != 0)): return previous_callback[0]
def allclose(net1: nn.Module, net2: nn.Module, /) -> bool: for (p1, p2) in zip(net1.parameters(), net2.parameters()): try: if (not p1.allclose(p2)): return False except RuntimeError: return False return True
def make_random_policy_rubiks_cube(rubiks_cube: RubiksCube) -> RandomPolicy: action_minimum = rubiks_cube.action_spec().minimum action_maximum = rubiks_cube.action_spec().maximum def random_policy(observation: Observation, key: chex.PRNGKey) -> chex.Array: batch_size = observation.cube.shape[0] action = jax.random.randint(key, (batch_size, len(action_minimum)), minval=action_minimum, maxval=action_maximum) return action return random_policy
class ResNet(nn.Module): def __init__(self, block, layers, num_classes, stop_layer): self.inplanes = 64 super(ResNet, self).__init__() self.stop = stop_layer self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(64, affine=affine_par) for i in self.bn1.parameters(): i.requires_grad = False self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1, ceil_mode=True) self.layer1 = self._make_layer(block, 64, layers[0]) self.layer2 = self._make_layer(block, 128, layers[1], stride=2) self.layer3 = self._make_layer(block, 256, layers[2], stride=1, dilation=2) if (not (stop_layer == 'layer4')): self.layer4 = self._make_layer(block, 512, layers[3], stride=1, dilation=4) for m in self.modules(): if isinstance(m, nn.Conv2d): n = ((m.kernel_size[0] * m.kernel_size[1]) * m.out_channels) m.weight.data.normal_(0, 0.01) elif isinstance(m, nn.BatchNorm2d): m.weight.data.fill_(1) m.bias.data.zero_() def _make_layer(self, block, planes, blocks, stride=1, dilation=1): downsample = None if ((stride != 1) or (self.inplanes != (planes * block.expansion)) or (dilation == 2) or (dilation == 4)): downsample = nn.Sequential(nn.Conv2d(self.inplanes, (planes * block.expansion), kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d((planes * block.expansion), affine=affine_par)) for i in downsample._modules['1'].parameters(): i.requires_grad = False layers = [] layers.append(block(self.inplanes, planes, stride, dilation=dilation, downsample=downsample)) self.inplanes = (planes * block.expansion) for i in range(1, blocks): layers.append(block(self.inplanes, planes, dilation=dilation)) return nn.Sequential(*layers) def _make_pred_layer(self, block, dilation_series, padding_series, num_classes): return block(dilation_series, padding_series, num_classes) def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) out1 = self.layer1(x) out2 = self.layer2(out1) out3 = self.layer3(out2) if (not (self.stop == 'layer4')): out4 = self.layer4(out3) outputs = [] outputs.append(out1) outputs.append(out2) outputs.append(out3) if (not (self.stop == 'layer4')): outputs.append(out4) return outputs
def text_is_range(tag_before_tag, tag, tag_after_tag): return ((tag_before_tag[1] == 'CD') and (tag[0] == '-') and (tag_after_tag[1] == 'CD'))
def validate(content, path): match = re.search('.*-([a-f0-9]{8,})\\.[a-zA-Z0-9]{2,}$', path) if match: stored_hsh = match.group(1) computed_hsh = hashlib.sha256(content).hexdigest()[:len(stored_hsh)] if (computed_hsh != stored_hsh): raise ValueError(("Computed hash '%s' is not consistent with stored hash '%s'" % (computed_hsh, stored_hsh)))
def avg_spatial_inner_product(a, b=None, batch_dims: int=0): _a = tf.reshape(a, (list(a.shape[:(- 3)]) + [(- 1), a.shape[(- 1)]])) if (b is None): _b = _a else: _b = tf.reshape(b, (list(b.shape[:(- 3)]) + [(- 1), b.shape[(- 1)]])) batch_axes = (2 * [list(range(batch_dims))]) prod = batch_tensordot(_a, _b, axes=[(- 1), (- 1)], batch_axes=batch_axes) return tf.reduce_mean(prod, [(- 3), (- 1)])
def initializer(in_filters, out_filters, name, k_size=3): w1 = tf.get_variable((name + 'W'), [k_size, k_size, in_filters, out_filters], initializer=tf.truncated_normal_initializer()) b1 = tf.get_variable((name + 'B'), [out_filters], initializer=tf.truncated_normal_initializer()) return (w1, b1)
class BackendNull(backend.Backend): def __init__(self): super(BackendNull, self).__init__() def version(self): return '-' def name(self): return 'null' def image_format(self): return 'NHWC' def load(self, model_path, inputs=None, outputs=None): self.outputs = ['output'] self.inputs = ['input'] return self def predict(self, feed): time.sleep(0) return [[0]]
def make_builder(out_file, impl, vocab_size=None): if (impl == 'mmap'): return MMapIndexedDatasetBuilder(out_file, dtype=best_fitting_int_dtype(vocab_size)) elif (impl == 'fasta'): raise NotImplementedError elif (impl == 'huffman'): raise ValueError('Use HuffmanCodeBuilder directly as it has a different interface.') else: return IndexedDatasetBuilder(out_file)
def read_samples(file_path, word2idx, whitelist): samples = [] with open(file_path, 'r') as f: for line in f: (task, story) = line.rstrip('\n').split('\t') if ((str(task) in whitelist.split('t')) or (len(whitelist) == 0)): words = story.split(' ') EOS = word2idx[END_OF_STORY] x = ([EOS] + [word2idx[word] for word in words]) y = ([word2idx[word] for word in words] + [EOS]) t = ([int(task)] * len(x)) samples.append((x, y, t)) return samples
def test_column_parallel_linear(model_parallel_size): mpu.initialize_model_parallel(model_parallel_size) if (torch.distributed.get_rank() == 0): print('> testing ColumnParallelLinear with model parallel size: {}'.format(model_parallel_size)) model_parallel_size = mpu.get_model_parallel_world_size() seed = 12345 set_random_seed(seed) input_size_coeff = 13 input_size = (input_size_coeff * model_parallel_size) output_size_coeff = 17 output_size = (output_size_coeff * model_parallel_size) batch_size = 7 identity_layer = IdentityLayer2D(batch_size, input_size).cuda() linear_layer = mpu.ColumnParallelLinear(input_size, output_size, keep_master_weight_for_test=True).cuda() loss_weight = torch.randn([batch_size, output_size]).cuda() input_ = identity_layer() output = linear_layer(input_) loss = torch.mul(output, loss_weight).sum() loss.backward() dLdY = loss_weight X = identity_layer.weight A = linear_layer.master_weight.cuda() dLdA = torch.matmul(dLdY.t(), X) dLdb = torch.matmul(torch.ones(batch_size, 1).cuda().t(), dLdY).view((- 1)) dLdX = torch.matmul(dLdY, A) rank = mpu.get_model_parallel_rank() my_dLdA = torch.split(dLdA, output_size_coeff, dim=0)[rank].contiguous().clone() error = my_dLdA.sub(linear_layer.weight.grad).abs().max() torch.distributed.barrier() print(' error in dLdA on global rank {}: {}'.format(torch.distributed.get_rank(), error)) assert (error < 1e-06) my_dLdb = torch.split(dLdb, output_size_coeff, dim=0)[rank].contiguous().clone() error = my_dLdb.sub(linear_layer.bias.grad).abs().max() torch.distributed.barrier() print(' error in dLdb on global rank {}: {}'.format(torch.distributed.get_rank(), error)) assert (error < 1e-06) error = dLdX.sub(identity_layer.weight.grad).abs().max() torch.distributed.barrier() print(' error in dLdX on global rank {}: {}'.format(torch.distributed.get_rank(), error)) assert (error < 1e-06) mpu.destroy_model_parallel() torch.distributed.barrier() if (torch.distributed.get_rank() == 0): print(' >> passed the test :-)')
def analyze_sentence(logit: List, inp_entropy: List, pred_dist: numpy.ndarray, input_doc, input_bigram, nucleus_filter: bool=True, top_p=0.9) -> List: viz_outputs = [] for (log, ent) in zip(logit, inp_entropy): viz_outputs.append('{0}_{1:.1f}'.format(bpe_tokenizer.decode(log), ent)) print(' '.join(viz_outputs)) cand_bigram = get_bigram(logit) l = len(logit) rt = [] return_pos = [] return_pos.append([0, l, comp_entropy(pred_dist[0], nucleus_filter, top_p)]) for (idx, big) in enumerate(cand_bigram): t = big[1][0] ent = comp_entropy(pred_dist[t], nucleus_filter, top_p) tok = big[1][2] (bigran, trigram) = (False, False) if (t >= 0): _can_big = f'{big[0][2]}_{big[1][2]}' if (_can_big in input_bigram): bigran = True else: bigran = False rt.append([t, l, ent, 0, tok, bigran, trigram]) return_pos.append([t, l, ent]) return (rt, return_pos)
def decompress_data(compressor, Z_bytes): start = time.time() with torch.no_grad(): Z_hat = [compressor.decompress(b).cpu().numpy() for b in Z_bytes] sec_per_img = ((time.time() - start) / len(Z_hat)) return (np.concatenate(Z_hat), sec_per_img)
def _test(): import torch pretrained = False models = [msdnet22_cifar10] for model in models: net = model(pretrained=pretrained) net.eval() weight_count = _calc_width(net) print('m={}, {}'.format(model.__name__, weight_count)) assert ((model != msdnet22_cifar10) or (weight_count == 4839544)) x = torch.randn(1, 3, 32, 32) y = net(x) y.sum().backward() assert (tuple(y.size()) == (1, 10))
class Agent(): def __init__(self, device_name, model_name, num_observations, num_envs, num_threads, training_data_path): print('Agent init !') self.num_envs = num_envs def pre_step(self, observations): actions = [] for i in range(self.num_envs): actions.append(random.randrange(0, NUM_ACTIONS)) self.last_action = actions[0] return actions def post_step(self, new_observations, rewards, dones, mean_reward, mean_success): return 0 def get_stats(self): return [('Action chosen', self.last_action), ('Pi', 3.14159)] def stop(self): pass
def log_logistic(x, mean, logscale): scale = torch.exp(logscale) u = ((x - mean) / scale) logp = ((F.logsigmoid(u) + F.logsigmoid((- u))) - logscale) return logp
class ONNXModel(TorchModel): def __init__(self, with_torch=True): super().__init__() self.with_torch: bool = with_torch self.masked_output_like: Optional[Dict[(str, AbsTensor)]] = None self.full_output_like: Optional[Dict[(str, AbsTensor)]] = None self.full_input_like: Optional[Dict[(str, AbsTensor)]] = None self.onnx_model = None def version(self) -> str: return f'onnx{onnx.__version__}-torch{torch.__version__}' def _mask_outputs(self) -> Dict[(str, AbsTensor)]: assert (self.torch_model is not None), 'Create a concrete model before masking outputs.' after_mask = {k: v for (k, v) in self.full_output_like.items() if (random.random() < 0.5)} if (len(after_mask) == 0): only_key = random.sample(self.full_output_like.keys(), 1)[0] after_mask = {only_key: self.full_output_like[only_key]} return after_mask def _dce_prob() -> float: dce_prob = 0.0 dce_env = os.getenv('NNSMITH_ONNX_DCE') if (dce_env is not None): if ((not dce_env.replace('.', '', 1).isdigit()) or (not (0 < float(dce_env) < 1))): raise ValueError(f'NNSMITH_ONNX_DCE must be [0, 1], but got {dce_env}') dce_prob = float(dce_env) return dce_prob def from_gir(cls: Type['ONNXModel'], gir: GraphIR, **kwargs) -> 'ONNXModel': ret = cls() ret.torch_model = SymbolNet(gir, **kwargs) ret.full_input_like = ret.torch_model.input_like ret.full_output_like = ret.torch_model.output_like ret.masked_output_like = ret.full_output_like if (random.random() < cls._dce_prob()): ret.masked_output_like = ret._mask_outputs() return ret def refine_weights(self) -> None: TorchModel.refine_weights(self) self.onnx_model = self.get_onnx_from_torch() if (set(self.masked_output_like.keys()) != set(self.full_output_like)): self.onnx_model = create_deadcode_onnx(self.onnx_model, self.masked_output_like.keys()) def input_like(self) -> Dict[(str, AbsTensor)]: return self.full_input_like def output_like(self) -> Dict[(str, AbsTensor)]: return self.masked_output_like def dump(self, path: PathLike) -> None: if self.with_torch: TorchModel.dump(self, path.replace(self.name_suffix(), TorchModel.name_suffix())) if (self.onnx_model is None): self.onnx_model = self.get_onnx_from_torch() onnx.checker.check_model(self.onnx_model, full_check=True) onnx.save(self.onnx_model, path) def load(cls, path: PathLike) -> 'ONNXModel': ret = cls() ret.onnx_model = onnx.load(path) torch_path = path.replace(cls.name_suffix(), TorchModel.name_suffix()) ret.with_torch = False (full_input_like, full_output_like) = analyze_onnx_io(ret.onnx_model) ret.full_input_like = full_input_like ret.full_output_like = full_output_like ret.masked_output_like = ret.full_output_like if os.path.exists(torch_path): ret.with_torch = True ret.torch_model = TorchModel.load(torch_path) ret.full_input_like = ret.torch_model.input_like ret.full_output_like = ret.torch_model.output_like return ret def native_model(self): if (self.with_torch and (self.onnx_model is None)): self.onnx_model = self.get_onnx_from_torch() return self.onnx_model def get_onnx_from_torch(self) -> onnx.ModelProto: f = BytesIO() torch2onnx(self.torch_model, f) onnx_model = onnx.load_model_from_string(f.getvalue()) onnx_model = update_model_dims.update_inputs_outputs_dims(onnx_model, {k: v.shape for (k, v) in self.torch_model.input_like.items()}, {k: v.shape for (k, v) in self.torch_model.output_like.items()}) onnx.checker.check_model(onnx_model, full_check=True) return onnx_model def name_suffix() -> str: return '.onnx'
class DiscourseParsingException(Exception): def __init__(self, message): self.message = message def __str__(self): return self.message
def get_device(batch: Any) -> Optional[torch.device]: if isinstance(batch, dict): return get_device(list(batch.values())) elif isinstance(batch, list): devices = [get_device(d) for d in batch] for d in devices: if (d is not None): return d else: return None elif isinstance(batch, torch.Tensor): return batch.device else: return None
def sparse_reward_function(state: State) -> chex.Array: solved = is_solved(state.cube) return jnp.array(solved, float)
def draw_bounding_box_on_image_array(image, ymin, xmin, ymax, xmax, color='red', thickness=4, display_str_list=(), use_normalized_coordinates=True): image_pil = Image.fromarray(np.uint8(image)).convert('RGB') draw_bounding_box_on_image(image_pil, ymin, xmin, ymax, xmax, color, thickness, display_str_list, use_normalized_coordinates) np.copyto(image, np.array(image_pil))
def main(): dataset = LinkPropPredDataset(name='ogbl-vessel') data = dataset[0] edge_index = data['edge_index'] print(f'Number of undirected training edges in the graph (accessing edge_index) :{edge_index.shape[1]}') print(f"Number of nodes in the graph:{data['num_nodes']}") print('Examining the splits') split = dataset.get_edge_split() split_train_pos = split['train']['edge'].cpu().detach().numpy() split_train_neg = split['train']['edge_neg'].cpu().detach().numpy() split_test_pos = split['test']['edge'].cpu().detach().numpy() split_test_neg = split['test']['edge_neg'].cpu().detach().numpy() split_valid_pos = split['valid']['edge'].cpu().detach().numpy() split_valid_neg = split['valid']['edge_neg'].cpu().detach().numpy() print(f'Dimensions of positive training edges: {split_train_pos.shape[0]}') print(f'Dimensions of positive test edges {split_test_pos.shape[0]}') print(f'Dimensions of positive validation edges {split_valid_pos.shape[0]}') print(f'Dimensions of negative training edges {split_train_neg.shape[0]}') print(f'Dimensions of negative test edges {split_test_neg.shape[0]}') print(f'Dimensions of negative validaztion edges {split_valid_neg.shape[0]}') (u, indices) = np.unique(split_train_pos, return_index=True, axis=0) print('Train edge pos - unique elements', len(u)) (u, indices) = np.unique(split_train_neg, return_index=True, axis=0) print('Train edge neg - unique elements', len(u)) (u, indices) = np.unique(split_valid_pos, return_index=True, axis=0) print('Valid edge pos - unique elements', len(u)) (u, indices) = np.unique(split_valid_neg, return_index=True, axis=0) print('Valid edge neg - unique elements', len(u)) (u, indices) = np.unique(split_test_pos, return_index=True, axis=0) print('Test edge pos - unique elements', len(u)) (u, indices) = np.unique(split_test_neg, return_index=True, axis=0) print('Test edge neg - unique elements', len(u)) print('') array = np.concatenate((split_train_pos, split_train_neg, split_test_pos, split_test_neg, split_valid_pos, split_valid_neg), axis=0) (u, indices) = np.unique(array, return_index=True, axis=0) print('All unique elements calculated', len(u)) print(f'All unique elements in theory: {(((split_train_pos.shape[0] * 2) + (split_valid_pos.shape[0] * 2)) + (split_test_pos.shape[0] * 2))}') print('Is the data representive - e.g. are test and validation set similar to training set') train_nodes = np.array(split_train_pos).flatten() valid_nodes = np.array(split_valid_pos).flatten() test_nodes = np.array(split_test_pos).flatten() plot_nodes(train_nodes, data, str='train') plot_nodes(valid_nodes, data, str='valid') plot_nodes(test_nodes, data, str='test') print('') train_pos_hist = plot_dist(split_train_pos, data, 'train_pos') valid_pos_hist = plot_dist(split_valid_pos, data, 'valid_pos') test_pos_hist = plot_dist(split_test_pos, data, 'test_pos') (stat, pvalue) = stats.ks_2samp(train_pos_hist, valid_pos_hist) print(f'KS-Test stats, p-value train vs valid set {(stat, pvalue)}') (stat, pvalue) = stats.ks_2samp(train_pos_hist, test_pos_hist) print(f'KS-Test stats, p-value test vs valid set {(stat, pvalue)}') (stat, pvalue) = stats.ks_2samp(valid_pos_hist, test_pos_hist) print(f'KS-Test stats, p-value valid vs test set {(stat, pvalue)}') train_neg_hist = plot_dist(split_train_neg, data, 'train_neg') valid_neg_hist = plot_dist(split_valid_neg, data, 'valid_neg') test_neg_hist = plot_dist(split_test_neg, data, 'test_neg') (stat, pvalue) = stats.ks_2samp(train_neg_hist, valid_neg_hist) print(f'KS-Test stats, p-value train neg vs valid neg set {(stat, pvalue)}') (stat, pvalue) = stats.ks_2samp(train_neg_hist, test_neg_hist) print(f'KS-Test stats, p-value train neg vs test neg set {(stat, pvalue)}') (stat, pvalue) = stats.ks_2samp(valid_neg_hist, test_neg_hist) print(f'KS-Test stats, p-value valid neg vs test neg set {(stat, pvalue)}')
def get_cifar10_mlp(): nb_classes = 10 batch_size = 64 epochs = 4 input_shape = (3072,) ((x_train, y_train), (x_test, y_test)) = cifar10.load_data() x_train = x_train.reshape(50000, 3072) x_test = x_test.reshape(10000, 3072) x_train = x_train.astype('float32') x_test = x_test.astype('float32') x_train /= 255 x_test /= 255 y_train = to_categorical(y_train, nb_classes) y_test = to_categorical(y_test, nb_classes) return (nb_classes, batch_size, input_shape, x_train, x_test, y_train, y_test, epochs)
class TestBenchmarkTanhNormalDistribution(): def test_new_tanh_normal(self): mean = torch.ones(1) std = torch.ones(1) dist = TanhNormal(mean, std) del dist def test_tanh_normal_bounds(self): mean = (torch.ones(1) * 100) std = (torch.ones(1) * 100) dist = TanhNormal(mean, std) assert (dist.mean <= 1.0) del dist mean = (torch.ones(1) * (- 100)) std = (torch.ones(1) * 100) dist = TanhNormal(mean, std) assert (dist.mean >= (- 1.0)) def test_tanh_normal_rsample(self): mean = torch.zeros(1) std = torch.ones(1) dist = TanhNormal(mean, std) sample = dist.rsample() (pre_tanh_action, action) = dist.rsample_with_pre_tanh_value() assert (pre_tanh_action.tanh() == action).all() assert ((- 1) <= action <= 1.0) assert ((- 1) <= sample <= 1.0) del dist def test_tanh_normal_log_prob(self): mean = torch.zeros(1) std = torch.ones(1) dist = TanhNormal(mean, std) pre_tanh_action = torch.Tensor([[2.096]]) action = pre_tanh_action.tanh() log_prob = dist.log_prob(action, pre_tanh_action) log_prob_approx = dist.log_prob(action) assert torch.allclose(log_prob, torch.Tensor([(- 0.2798519)])) assert torch.allclose(log_prob_approx, torch.Tensor([(- 0.2798519)])) del dist def test_tanh_normal_expand(self): mean = torch.zeros(1) std = torch.ones(1) dist = TanhNormal(mean, std) new_dist = dist.expand((2,)) sample = new_dist.sample() assert (sample.shape == torch.Size((2, 1))) def test_tanh_normal_repr(self): mean = torch.zeros(1) std = torch.ones(1) dist = TanhNormal(mean, std) assert (repr(dist) == 'TanhNormal')
def create_zip(data_root: Path, zip_path: Path): paths = list(data_root.glob('*.npy')) with zipfile.ZipFile(zip_path, 'w', zipfile.ZIP_STORED) as f: for path in tqdm(paths): f.write(path, arcname=path.name)
class ShmemVecEnv(VecEnv): def __init__(self, env_fns, spaces=None): if spaces: (observation_space, action_space) = spaces else: logger.log('Creating dummy env object to get spaces') with logger.scoped_configure(format_strs=[]): dummy = env_fns[0]() (observation_space, action_space) = (dummy.observation_space, dummy.action_space) dummy.close() del dummy VecEnv.__init__(self, len(env_fns), observation_space, action_space) (self.obs_keys, self.obs_shapes, self.obs_dtypes) = obs_space_info(observation_space) self.obs_bufs = [{k: Array(_NP_TO_CT[self.obs_dtypes[k].type], int(np.prod(self.obs_shapes[k]))) for k in self.obs_keys} for _ in env_fns] self.parent_pipes = [] self.procs = [] for (env_fn, obs_buf) in zip(env_fns, self.obs_bufs): wrapped_fn = CloudpickleWrapper(env_fn) (parent_pipe, child_pipe) = Pipe() proc = Process(target=_subproc_worker, args=(child_pipe, parent_pipe, wrapped_fn, obs_buf, self.obs_shapes, self.obs_dtypes, self.obs_keys)) proc.daemon = True self.procs.append(proc) self.parent_pipes.append(parent_pipe) proc.start() child_pipe.close() self.waiting_step = False self.viewer = None def reset(self): if self.waiting_step: logger.warn('Called reset() while waiting for the step to complete') self.step_wait() for pipe in self.parent_pipes: pipe.send(('reset', None)) return self._decode_obses([pipe.recv() for pipe in self.parent_pipes]) def step_async(self, actions): assert (len(actions) == len(self.parent_pipes)) for (pipe, act) in zip(self.parent_pipes, actions): pipe.send(('step', act)) def step_wait(self): outs = [pipe.recv() for pipe in self.parent_pipes] (obs, rews, dones, infos) = zip(*outs) return (self._decode_obses(obs), np.array(rews), np.array(dones), infos) def close_extras(self): if self.waiting_step: self.step_wait() for pipe in self.parent_pipes: pipe.send(('close', None)) for pipe in self.parent_pipes: pipe.recv() pipe.close() for proc in self.procs: proc.join() def get_images(self, mode='human'): for pipe in self.parent_pipes: pipe.send(('render', None)) return [pipe.recv() for pipe in self.parent_pipes] def _decode_obses(self, obs): result = {} for k in self.obs_keys: bufs = [b[k] for b in self.obs_bufs] o = [np.frombuffer(b.get_obj(), dtype=self.obs_dtypes[k]).reshape(self.obs_shapes[k]) for b in bufs] result[k] = np.array(o) return dict_to_obs(result)
def Conv3D(inputs, f_dim_out, net, f_dim_in=None, batch_norm=False, is_train=True): if (f_dim_in is None): f_dim_in = int((f_dim_out / 2)) layer = tl.layers.Conv3dLayer(inputs, shape=[4, 4, 4, f_dim_in, f_dim_out], W_init=tf.random_normal_initializer(stddev=0.02), strides=[1, 2, 2, 2, 1], name=(('d/net_' + net) + '/conv')) if batch_norm: return tl.layers.BatchNormLayer(layer, is_train=is_train, name=(('d/net_' + net) + '/batch_norm')) else: return layer
class BackendTensorflow(backend.Backend): def __init__(self): super(BackendTensorflow, self).__init__() def version(self): return ((tf.__version__ + '/') + tf.__git_version__) def name(self): return 'tensorflow' def image_format(self): return 'NHWC' def load(self, model_path, inputs=None, outputs=None): if (not inputs): raise ValueError('BackendTensorflow needs inputs') if (not outputs): raise ValueError('BackendTensorflow needs outputs') self.outputs = outputs self.inputs = inputs infer_config = tf.compat.v1.ConfigProto() infer_config.intra_op_parallelism_threads = (int(os.environ['TF_INTRA_OP_PARALLELISM_THREADS']) if ('TF_INTRA_OP_PARALLELISM_THREADS' in os.environ) else os.cpu_count()) infer_config.inter_op_parallelism_threads = (int(os.environ['TF_INTER_OP_PARALLELISM_THREADS']) if ('TF_INTER_OP_PARALLELISM_THREADS' in os.environ) else os.cpu_count()) infer_config.use_per_session_threads = 1 graph_def = tf.compat.v1.GraphDef() with tf.compat.v1.gfile.FastGFile(model_path, 'rb') as f: graph_def.ParseFromString(f.read()) for as_datatype_enum in [dtypes.float32.as_datatype_enum, dtypes.uint8.as_datatype_enum]: try: optimized_graph_def = optimize_for_inference(graph_def, [item.split(':')[0] for item in inputs], [item.split(':')[0] for item in outputs], as_datatype_enum, False) graph_def = optimized_graph_def break except ValueError: pass g = tf.compat.v1.import_graph_def(graph_def, name='') self.sess = tf.compat.v1.Session(graph=g, config=infer_config) return self def predict(self, feed): return self.sess.run(self.outputs, feed_dict=feed)
_module() class SemiLossCPS(nn.Module): def __init__(self, loss_weight=1.0, avg_non_ignore=True, ignore_index=255): super(SemiLossCPS, self).__init__() self.loss_weight = loss_weight self.ignore_index = ignore_index self.criterion = CrossEntropyLoss(loss_weight=loss_weight, avg_non_ignore=avg_non_ignore) def forward(self, strong_logits, weak_logits): (max_probs, targets_u) = torch.max(weak_logits, dim=1) loss1 = self.criterion(strong_logits.float(), targets_u.long(), ignore_index=self.ignore_index) (max_probs2, targets_u2) = torch.max(strong_logits, dim=1) loss2 = self.criterion(weak_logits.float(), targets_u2.long(), ignore_index=self.ignore_index) loss = (loss1 + loss2) return loss
_model def seresnext101_32x4d(pretrained=False, num_classes=1000, in_chans=3, **kwargs): default_cfg = default_cfgs['seresnext101_32x4d'] model = SENet(SEResNeXtBottleneck, [3, 4, 23, 3], groups=32, reduction=16, inplanes=64, input_3x3=False, downsample_kernel_size=1, downsample_padding=0, num_classes=num_classes, in_chans=in_chans, **kwargs) model.default_cfg = default_cfg if pretrained: load_pretrained(model, default_cfg, num_classes, in_chans) return model
def assign_id_to_src(src_name, src_value): if isinstance(src_value, six.string_types): src_id = src_value else: try: src_id = src_value.id except: err = '{0} does not have an `id` property. {1} needs to be assigned to either an object with an `id` (like a plotly.grid_objs.Column) or a string. The `id` is a unique identifier assigned by the Plotly webserver to this grid column.' src_value_str = str(src_value) err = err.format(src_name, src_value_str) raise exceptions.InputError(err) if (src_id == ''): err = exceptions.COLUMN_NOT_YET_UPLOADED_MESSAGE err.format(column_name=src_value.name, reference=src_name) raise exceptions.InputError(err) return src_id