def main(): # Get data (train_images, train_labels), (test_images, test_labels) = keras.datasets.fashion_mnist.load_data() train_images = train_images.astype(np.float32) / 255.0 test_images = test_images.astype(np.float32) / 255.0 # Load saved model model = tf.keras.models.load_model("../models/simple_model") # Run ilit to get the quantized graph tuner = ilit.Tuner('./conf.yaml') dataloader = tuner.dataloader(dataset=(test_images, test_labels)) quantized_model = tuner.tune(model, q_dataloader=dataloader, eval_func=eval_func) # Run inference with quantized model concrete_function = get_concrete_function( graph_def=quantized_model.as_graph_def(), inputs=["args_0:0"], outputs=["Identity:0"], print_graph=True) frozen_graph_predictions = concrete_function( args_0=tf.constant(test_images))[0] print("Inference is done.")
def main(): # Get data (train_images, train_labels), (test_images, test_labels) = keras.datasets.fashion_mnist.load_data() train_images = train_images.astype(np.float32) / 255.0 test_images = test_images.astype(np.float32) / 255.0 # Load model model_file = "../frozen_models/simple_frozen_graph.pb" graph = load_graph(model_file) # Run ilit to get the quantized pb tuner = ilit.Tuner('./conf.yaml') dataloader = tuner.dataloader(dataset=(test_images, test_labels)) quantized_model = tuner.tune(graph, q_dataloader=dataloader, eval_func=eval_func) # Run quantized model with tf.compat.v1.Session() as sess: tf.compat.v1.import_graph_def(quantized_model.as_graph_def()) styled_image = sess.run(['import/Identity:0'], feed_dict={'import/x:0': test_images}) print("Inference is done.")
def auto_tune(input_graph_path, yaml_config, batch_size): fp32_graph = alexnet.load_pb(input_graph_path) tuner = ilit.Tuner(yaml_config) dataloader = Dataloader(batch_size) q_model = tuner.tune( fp32_graph, q_dataloader=dataloader, eval_func=None, eval_dataloader=dataloader) return q_model
def auto_tune(self): """This is ilit tuning part to generate a quantized pb Returns: graph: it will return a quantized pb """ import ilit fp32_graph = load_graph(self.args.input_graph) tuner = ilit.Tuner(self.args.config) if self.args.calib_data: calib_dataloader = Dataloader(self.args.calib_data, self.args.batch_size) q_model = tuner.tune(fp32_graph, q_dataloader=calib_dataloader, eval_func=self.eval_inference, eval_dataloader=None) return q_model else: print("Please provide calibration dataset!")
def auto_tune(self): """This is iLiT tuning part to generate a quantized pb Returns: graph: it will return a quantized pb """ import ilit fp32_graph = load_graph(self.args.input_graph) tuner = ilit.Tuner(self.args.config) dataloader = Dataloader(self.args.data_location, 'validation', self.args.image_size, self.args.image_size, self.args.batch_size, self.args.num_cores, self.args.resize_method, [self.args.r_mean,self.args.g_mean,self.args.b_mean], self.args.label_adjust) q_model = tuner.tune( fp32_graph, q_dataloader=dataloader, # eval_func=self.eval_inference) eval_func=None, eval_dataloader=dataloader) return q_model
def main(): # init the args args = Options().parse() args.cuda = not args.no_cuda and torch.cuda.is_available() print(args) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) # init dataloader interp = PIL.Image.BILINEAR if args.crop_size < 320 else PIL.Image.BICUBIC base_size = args.base_size if args.base_size is not None else int( 1.0 * args.crop_size / 0.875) transform_val = transforms.Compose([ ECenterCrop(args.crop_size), transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]), ]) valset = ImageNetDataset(args.data, transform=transform_val, train=False) val_loader = torch.utils.data.DataLoader( valset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers, pin_memory=True if args.cuda else False) # assert args.model in torch.hub.list('zhanghang1989/ResNeSt', force_reload=True) functions = inspect.getmembers(module, inspect.isfunction) model_list = [f[0] for f in functions] assert args.model in model_list get_model = importlib.import_module('resnest.torch') net = getattr(get_model, args.model) # model = torch.hub.load('zhanghang1989/ResNeSt', args.model, pretrained=True) model = net(pretrained=True) # print(model) if args.cuda: model.cuda() # Please use CUDA_VISIBLE_DEVICES to control the number of gpus model = nn.DataParallel(model) # checkpoint if args.verify: if os.path.isfile(args.verify): print("=> loading checkpoint '{}'".format(args.verify)) model.module.load_state_dict(torch.load(args.verify)) else: raise RuntimeError("=> no verify checkpoint found at '{}'".format( args.verify)) elif args.resume is not None: if os.path.isfile(args.resume): print("=> loading checkpoint '{}'".format(args.resume)) checkpoint = torch.load(args.resume) model.module.load_state_dict(checkpoint['state_dict']) else: raise RuntimeError("=> no resume checkpoint found at '{}'".format( args.resume)) model.eval() if args.tune: model.fuse_model() import ilit tuner = ilit.Tuner("./conf.yaml") q_model = tuner.tune(model) exit(0) top1 = AverageMeter() top5 = AverageMeter() batch_time = AverageMeter() iterations = args.iterations warmup = args.warmup_iterations tbar = tqdm(val_loader, desc='\r') for batch_idx, (data, target) in enumerate(tbar): if iterations == 0 or batch_idx < iterations + warmup: if batch_idx >= warmup: end = time.time() if args.cuda: data, target = data.cuda(), target.cuda() with torch.no_grad(): output = model(data) if batch_idx >= warmup: batch_time.update(time.time() - end) acc1, acc5 = accuracy(output, target, topk=(1, 5)) top1.update(acc1[0], data.size(0)) top5.update(acc5[0], data.size(0)) tbar.set_description('Top1: %.3f | Top5: %.3f' % (top1.avg, top5.avg)) elif batch_idx == iterations + warmup: break print('Top1 Acc: %.3f | Top5 Acc: %.3f ' % (top1.avg, top5.avg))
def main(args=None): tf.logging.set_verbosity(tf.logging.INFO) if not tf.gfile.Exists(FLAGS.output_dir): tf.gfile.MkDir(FLAGS.output_dir) with tf.Session() as sess: if FLAGS.input_model.rsplit('.', 1)[-1] == 'ckpt': style_img_ph = tf.placeholder(tf.float32, shape=[None, 256, 256, 3], name='style_input') content_img_ph = tf.placeholder(tf.float32, shape=[None, 256, 256, 3], name='content_input') # import meta_graph meta_data_path = FLAGS.input_model + '.meta' saver = tf.train.import_meta_graph(meta_data_path, clear_devices=True) sess.run(tf.global_variables_initializer()) saver.restore(sess, FLAGS.input_model) graph_def = sess.graph.as_graph_def() replace_style = 'style_image_processing/ResizeBilinear_2' replace_content = 'batch_processing/batch' for node in graph_def.node: for idx, input_name in enumerate(node.input): # replace style input and content input nodes to placeholder if replace_content == input_name: node.input[idx] = 'content_input' if replace_style == input_name: node.input[idx] = 'style_input' if FLAGS.tune: _parse_ckpt_bn_input(graph_def) output_name = 'transformer/expand/conv3/conv/Sigmoid' frozen_graph = tf.graph_util.convert_variables_to_constants(sess, graph_def, [output_name]) # use frozen pb instead elif FLAGS.input_model.rsplit('.', 1)[-1] == 'pb': with open(FLAGS.input_model, 'rb') as f: frozen_graph = tf.GraphDef() frozen_graph.ParseFromString(f.read()) else: print("not supported model format") exit(-1) if FLAGS.tune: with tf.Graph().as_default() as graph: tf.import_graph_def(frozen_graph) tuner = ilit.Tuner(FLAGS.config) quantized_model = tuner.tune(graph, eval_func=eval_func) # save the frozen model for deployment with tf.io.gfile.GFile(FLAGS.output_model, "wb") as f: f.write(quantized_model.as_graph_def().SerializeToString()) frozen_graph= quantized_model.as_graph_def() # validate the quantized model here with tf.Graph().as_default(), tf.Session() as sess: # create dataloader using default style_transfer dataset and generate stylized images dataset = ilit.data.DATASETS('tensorflow')['style_transfer'](FLAGS.content_images_paths, FLAGS.style_images_paths, crop_ratio=0.2, resize_shape=(256, 256)) dataloader = ilit.data.DataLoader('tensorflow', dataset=dataset) tf.import_graph_def(frozen_graph) style_transfer(sess, dataloader, FLAGS.precision)
def main(): args = parser.parse_args() print(args) if args.img_size is None: args.img_size, args.crop_pct = get_image_size_crop_pct(args.model) if not args.checkpoint and not args.pretrained: args.pretrained = True if args.torchscript: geffnet.config.set_scriptable(True) # create model model = geffnet.create_model(args.model, num_classes=args.num_classes, in_chans=3, pretrained=args.pretrained, checkpoint_path=args.checkpoint) if args.torchscript: torch.jit.optimized_execution(True) model = torch.jit.script(model) print('Model %s created, param count: %d' % (args.model, sum([m.numel() for m in model.parameters()]))) data_config = resolve_data_config(model, args) criterion = nn.CrossEntropyLoss() if not args.no_cuda: if args.num_gpu > 1: model = torch.nn.DataParallel(model, device_ids=list(range( args.num_gpu))).cuda() else: model = model.cuda() criterion = criterion.cuda() if args.tune: model.eval() model.fuse_model() conf_yaml = "conf_" + args.model + ".yaml" import ilit tuner = ilit.Tuner(conf_yaml) q_model = tuner.tune(model) exit(0) valdir = os.path.join(args.data, 'val') loader = create_loader(Dataset(valdir, load_bytes=args.tf_preprocessing), input_size=data_config['input_size'], batch_size=args.batch_size, use_prefetcher=not args.no_cuda, interpolation=data_config['interpolation'], mean=data_config['mean'], std=data_config['std'], num_workers=args.workers, crop_pct=data_config['crop_pct'], tensorflow_preprocessing=args.tf_preprocessing) batch_time = AverageMeter() losses = AverageMeter() top1 = AverageMeter() top5 = AverageMeter() model.eval() end = time.time() with torch.no_grad(): for i, (input, target) in enumerate(loader): if not args.no_cuda: target = target.cuda() input = input.cuda() # compute output output = model(input) loss = criterion(output, target) # measure accuracy and record loss prec1, prec5 = accuracy(output.data, target, topk=(1, 5)) losses.update(loss.item(), input.size(0)) top1.update(prec1.item(), input.size(0)) top5.update(prec5.item(), input.size(0)) # measure elapsed time batch_time.update(time.time() - end) end = time.time() if i % args.print_freq == 0: print( 'Test: [{0}/{1}]\t' 'Time {batch_time.val:.3f} ({batch_time.avg:.3f}, {rate_avg:.3f}/s) \t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format( i, len(loader), batch_time=batch_time, rate_avg=input.size(0) / batch_time.avg, loss=losses, top1=top1, top5=top5))
logger.info("%s is %.4f." % (metric_nm[0], acc)) logger.info("%s is %.4f." % (metric_nm[1], F1)) logger.info("speed is %.2f samples/s" % speed) return acc if __name__ == '__main__': if only_calibration: try: calibration(model, dev_data_list, num_calib_batches, quantized_dtype, calib_mode) except AttributeError: nlp.utils.version.check_version('1.7.0', warning_only=True, library=mx) warnings.warn( 'INT8 Quantization for BERT need mxnet-mkl >= 1.6.0b20200115') elif args.tune: # ilit auto-tuning if only_inference: calib_data = dev_data_list[0][1] import ilit bert_tuner = ilit.Tuner("./bert.yaml") bert_tuner.tune(model, q_dataloader=calib_data, eval_dataloader=calib_data, eval_func=test_func) else: train(task.metrics)
def main_worker(gpu, ngpus_per_node, args): global best_acc1 #args.gpu = gpu #affinity = subprocess.check_output("lscpu | grep 'NUMA node[0-9]' | awk '{ print $4 }' | awk -F',' '{ print $1 }'", shell=True) #os.environ['OMP_NUM_THREADS'] = '28' #os.environ['KMP_AFFINITY'] = 'proclist=[{}],granularity=thread,explicit'.format(affinity.splitlines()[gpu].decode('utf-8')) #print (os.environ['KMP_AFFINITY']) #if args.gpu is not None: # print("Use GPU: {} for training".format(args.gpu)) print("Use CPU: {} for training".format(gpu)) if args.distributed: if args.dist_url == "env://" and args.rank == -1: args.rank = int(os.environ["RANK"]) if args.multiprocessing_distributed: # For multiprocessing distributed training, rank needs to be the # global rank among all the processes args.rank = args.rank * ngpus_per_node + gpu dist.init_process_group(backend=args.dist_backend, init_method=args.dist_url, world_size=args.world_size, rank=args.rank) # create model if args.pretrained: print("=> using pre-trained model '{}'".format(args.arch)) model = models.__dict__[args.arch](pretrained=True, quantize=False) else: print("=> creating model '{}'".format(args.arch)) model = models.__dict__[args.arch]() if args.distributed: # For multiprocessing distributed, DistributedDataParallel constructor # should always set the single device scope, otherwise, # DistributedDataParallel will use all available devices. if args.gpu is not None: torch.cuda.set_device(args.gpu) model.cuda(args.gpu) # When using a single GPU per process and per # DistributedDataParallel, we need to divide the batch size # ourselves based on the total number of GPUs we have args.batch_size = int(args.batch_size / ngpus_per_node) args.workers = int(args.workers / ngpus_per_node) model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.gpu]) else: #model.cuda() # DistributedDataParallel will divide and allocate batch_size to all # available GPUs if device_ids are not set model = torch.nn.parallel.DistributedDataParallelCPU(model) elif args.gpu is not None: torch.cuda.set_device(args.gpu) model = model.cuda(args.gpu) else: # DataParallel will divide and allocate batch_size to all available GPUs if args.arch.startswith('alexnet') or args.arch.startswith('vgg'): model.features = torch.nn.DataParallel(model.features) model.cuda() else: model = torch.nn.DataParallel(model) # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss() #criterion = nn.CrossEntropyLoss().cuda(args.gpu) optimizer = torch.optim.SGD(model.parameters(), args.lr, momentum=args.momentum, weight_decay=args.weight_decay) # optionally resume from a checkpoint if args.resume: if os.path.isfile(args.resume): print("=> loading checkpoint '{}'".format(args.resume)) checkpoint = torch.load(args.resume) args.start_epoch = checkpoint['epoch'] best_acc1 = checkpoint['best_acc1'] if args.gpu is not None: # best_acc1 may be from a checkpoint from a different GPU best_acc1 = best_acc1.to(args.gpu) model.load_state_dict(checkpoint['state_dict']) optimizer.load_state_dict(checkpoint['optimizer']) print("=> loaded checkpoint '{}' (epoch {})".format( args.resume, checkpoint['epoch'])) else: print("=> no checkpoint found at '{}'".format(args.resume)) #cudnn.benchmark = True # Data loading code traindir = os.path.join(args.data, 'train') valdir = os.path.join(args.data, 'val') normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) train_dataset = datasets.ImageFolder( traindir, transforms.Compose([ transforms.RandomResizedCrop(224), transforms.RandomHorizontalFlip(), transforms.ToTensor(), normalize, ])) if args.distributed: train_sampler = torch.utils.data.distributed.DistributedSampler( train_dataset) else: train_sampler = None train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None), num_workers=args.workers, pin_memory=True, sampler=train_sampler) val_loader = torch.utils.data.DataLoader(datasets.ImageFolder( valdir, transforms.Compose([ transforms.Resize(256), transforms.CenterCrop(224), transforms.ToTensor(), normalize, ])), batch_size=args.batch_size, shuffle=False, num_workers=args.workers, pin_memory=True) if args.evaluate: validate(val_loader, model, criterion, args) if args.tune: model.eval() model.module.fuse_model() import ilit tuner = ilit.Tuner("./conf.yaml") q_model = tuner.tune(model) return for epoch in range(args.start_epoch, args.epochs): if args.distributed: train_sampler.set_epoch(epoch) adjust_learning_rate(optimizer, epoch, args) # train for one epoch train(train_loader, model, criterion, optimizer, epoch, args) # evaluate on validation set acc1 = validate(val_loader, model, criterion, args) # remember best acc@1 and save checkpoint is_best = acc1 > best_acc1 best_acc1 = max(acc1, best_acc1) if not args.multiprocessing_distributed or ( args.multiprocessing_distributed and args.rank % ngpus_per_node == 0): save_checkpoint( { 'epoch': epoch + 1, 'arch': args.arch, 'state_dict': model.state_dict(), 'best_acc1': best_acc1, 'optimizer': optimizer.state_dict(), }, is_best)
return AP.mean() model.eval() model.fuse_model() import ilit dataset = ListDataset(valid_path, img_size=opt.img_size, augment=False, multiscale=False) dataloader = torch.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=False, num_workers=1, collate_fn=dataset.collate_fn) ilit_dataloader = yolo_dataLoader(dataloader) tuner = ilit.Tuner("./conf.yaml") tuner.tune(model, q_dataloader=ilit_dataloader, eval_func=eval_func) exit(0) print("Compute mAP...") precision, recall, AP, f1, ap_class = evaluate( model, path=valid_path, iou_thres=opt.iou_thres, conf_thres=opt.conf_thres, nms_thres=opt.nms_thres, img_size=opt.img_size, batch_size=8, )
def main(): global args, best_prec1 args = parser.parse_args() # create model print("=> creating model '{}'".format(args.arch)) if args.pretrained.lower() not in ['false', 'none', 'not', 'no', '0']: print("=> using pre-trained parameters '{}'".format(args.pretrained)) model = pretrainedmodels.__dict__[args.arch]( num_classes=1000, pretrained=args.pretrained) else: model = pretrainedmodels.__dict__[args.arch]() # optionally resume from a checkpoint if args.resume: if os.path.isfile(args.resume): print("=> loading checkpoint '{}'".format(args.resume)) checkpoint = torch.load(args.resume) args.start_epoch = checkpoint['epoch'] best_prec1 = checkpoint['best_prec1'] model.load_state_dict(checkpoint['state_dict']) print("=> loaded checkpoint '{}' (epoch {})".format( args.resume, checkpoint['epoch'])) else: print("=> no checkpoint found at '{}'".format(args.resume)) cudnn.benchmark = True # Data loading code # traindir = os.path.join(args.data, 'train') valdir = os.path.join(args.data, 'val') # train_loader = torch.utils.data.DataLoader( # datasets.ImageFolder(traindir, transforms.Compose([ # transforms.RandomSizedCrop(max(model.input_size)), # transforms.RandomHorizontalFlip(), # transforms.ToTensor(), # normalize, # ])), # batch_size=args.batch_size, shuffle=True, # num_workers=args.workers, pin_memory=True) # if 'scale' in pretrainedmodels.pretrained_settings[args.arch][args.pretrained]: # scale = pretrainedmodels.pretrained_settings[args.arch][args.pretrained]['scale'] # else: # scale = 0.875 scale = 0.875 print('Images transformed from size {} to {}'.format( int(round(max(model.input_size) / scale)), model.input_size)) val_tf = pretrainedmodels.utils.TransformImage( model, scale=scale, preserve_aspect_ratio=args.preserve_aspect_ratio) val_loader = torch.utils.data.DataLoader(datasets.ImageFolder( valdir, val_tf), batch_size=args.batch_size, shuffle=False, num_workers=args.workers, pin_memory=True) # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.SGD(model.parameters(), args.lr, momentum=args.momentum, weight_decay=args.weight_decay) model = torch.nn.DataParallel(model) if args.tune: model.eval() model.module.fuse_model() import ilit tuner = ilit.Tuner("./conf.yaml") q_model = tuner.tune(model) return if args.evaluate: validate(val_loader, model, criterion) return for epoch in range(args.start_epoch, args.epochs): adjust_learning_rate(optimizer, epoch) # train for one epoch train(train_loader, model, criterion, optimizer, epoch) # evaluate on validation set prec1 = validate(val_loader, model, criterion) # remember best prec@1 and save checkpoint is_best = prec1 > best_prec1 best_prec1 = max(prec1, best_prec1) save_checkpoint( { 'epoch': epoch + 1, 'arch': args.arch, 'state_dict': model.state_dict(), 'best_prec1': best_prec1, }, is_best)
prefix = os.path.join(dst_dir, net_name + '-quantized-' + args.calib_mode) logger.info('Saving quantized model at %s' % dst_dir) net.export(prefix, epoch=0) sys.exit() def eval_func(graph): val_dataset, val_metric = get_dataset(args.dataset, args.data_shape) val_data = get_dataloader( val_dataset, args.data_shape, args.batch_size, args.num_workers) classes = val_dataset.classes # class names size = len(val_dataset) ctx = [mx.cpu()] results = validate(graph, val_data, ctx, classes, size, val_metric) mAP = float(results[-1][-1]) return mAP # Doing iLiT auto-tuning here import ilit ssd_tuner = ilit.Tuner("./ssd.yaml") ssd_tuner.tune(net, q_dataloader=val_data, eval_dataloader=val_dataset, eval_func=eval_func) sys.exit() # eval names, values = validate(net, val_data, ctx, classes, len(val_dataset), val_metric) for k, v in zip(names, values): print(k, v)
def main(): parser = argparse.ArgumentParser() ## Required parameters parser.add_argument("--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys())) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS)) parser.add_argument( "--output_dir", default=None, type=str, required=True, help= "The output directory where the model checkpoints and predictions will be written." ) ## Other parameters parser.add_argument( "--data_dir", default=None, type=str, help= "The input data dir. Should contain the .json files for the task. If not specified, will run with tensorflow_datasets." ) parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name") parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name") parser.add_argument( "--cache_dir", default="", type=str, help= "Where do you want to store the pre-trained models downloaded from s3") parser.add_argument( '--version_2_with_negative', action='store_true', help= 'If true, the SQuAD examples contain some that do not have an answer.') parser.add_argument( '--null_score_diff_threshold', type=float, default=0.0, help= "If null_score - best_non_null is greater than the threshold predict null." ) parser.add_argument( "--max_seq_length", default=384, type=int, help= "The maximum total input sequence length after WordPiece tokenization. Sequences " "longer than this will be truncated, and sequences shorter than this will be padded." ) parser.add_argument( "--doc_stride", default=128, type=int, help= "When splitting up a long document into chunks, how much stride to take between chunks." ) parser.add_argument( "--max_query_length", default=64, type=int, help= "The maximum number of tokens for the question. Questions longer than this will " "be truncated to this length.") parser.add_argument("--do_train", action='store_true', help="Whether to run training.") parser.add_argument("--do_eval", action='store_true', help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action='store_true', help="Rul evaluation during training at each logging step.") parser.add_argument( "--do_lower_case", action='store_true', help="Set this flag if you are using an uncased model.") parser.add_argument("--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.") parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.") parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument( '--gradient_accumulation_steps', type=int, default=1, help= "Number of updates steps to accumulate before performing a backward/update pass." ) parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument("--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.") parser.add_argument( "--max_steps", default=-1, type=int, help= "If > 0: set total number of training steps to perform. Override num_train_epochs." ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument( "--n_best_size", default=20, type=int, help= "The total number of n-best predictions to generate in the nbest_predictions.json output file." ) parser.add_argument( "--max_answer_length", default=30, type=int, help= "The maximum length of an answer that can be generated. This is needed because the start " "and end predictions are not conditioned on one another.") parser.add_argument( "--verbose_logging", action='store_true', help= "If true, all of the warnings related to data processing will be printed. " "A number of warnings are expected for a normal SQuAD evaluation.") parser.add_argument('--logging_steps', type=int, default=50, help="Log every X updates steps.") parser.add_argument('--save_steps', type=int, default=50, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action='store_true', help= "Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number" ) parser.add_argument("--no_cuda", action='store_true', help="Whether not to use CUDA when available") parser.add_argument('--overwrite_output_dir', action='store_true', help="Overwrite the content of the output directory") parser.add_argument( '--overwrite_cache', action='store_true', help="Overwrite the cached training and evaluation sets") parser.add_argument('--seed', type=int, default=42, help="random seed for initialization") parser.add_argument("--local_rank", type=int, default=-1, help="local_rank for distributed training on gpus") parser.add_argument( '--fp16', action='store_true', help= "Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit" ) parser.add_argument( '--fp16_opt_level', type=str, default='O1', help= "For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html") parser.add_argument('--server_ip', type=str, default='', help="Can be used for distant debugging.") parser.add_argument('--server_port', type=str, default='', help="Can be used for distant debugging.") parser.add_argument("--do_calibration", action='store_true', help="Whether to do calibration.") parser.add_argument("--do_int8_inference", action='store_true', help="Whether to run int8 inference.") parser.add_argument("--do_fp32_inference", action='store_true', help="Whether to run fp32 inference.") parser.add_argument("--mkldnn_eval", action='store_true', help="evaluation with MKLDNN") parser.add_argument("--do_ilit_tune", action='store_true', help="run iLiT tool to tune int8 acc.") parser.add_argument("--task_name", default=None, type=str, required=True, help="SQuAD task") parser.add_argument("--warmup", type=int, default=5, help="warmup for performance") args = parser.parse_args() args.predict_file = os.path.join( args.output_dir, 'predictions_{}_{}.txt'.format( list(filter(None, args.model_name_or_path.split('/'))).pop(), str(args.max_seq_length))) if os.path.exists(args.output_dir) and os.listdir( args.output_dir ) and args.do_train and not args.overwrite_output_dir: raise ValueError( "Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome." .format(args.output_dir)) mix_qkv = False if args.do_calibration or args.do_int8_inference or args.do_ilit_tune: mix_qkv = True # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend='nccl') args.n_gpu = 1 args.device = device # Setup logging logging.basicConfig( format='%(asctime)s - %(levelname)s - %(name)s - %(message)s', datefmt='%m/%d/%Y %H:%M:%S', level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN) logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16) # Set seed set_seed(args) # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier( ) # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, cache_dir=args.cache_dir if args.cache_dir else None) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool('.ckpt' in args.model_name_or_path), config=config, mix_qkv=mix_qkv, cache_dir=args.cache_dir if args.cache_dir else None) if args.local_rank == 0: torch.distributed.barrier( ) # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Before we do anything with models, we want to ensure that we get fp16 execution of torch.einsum if args.fp16 is set. # Otherwise it'll default to "promote" mode, and we'll get fp32 operations. Note that running `--fp16_opt_level="O2"` will # remove the need for this code, but it is still valid. if args.fp16: try: import apex apex.amp.register_half_function(torch, 'einsum') except ImportError: raise ImportError( "Please install apex from https://www.github.com/nvidia/apex to use fp16 training." ) # Training if args.do_train: train_dataset = load_and_cache_examples(args, tokenizer, evaluate=False, output_examples=False) global_step, tr_loss = train(args, train_dataset, model, tokenizer) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) # Save the trained model and the tokenizer if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0): # Create output directory if needed if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]: os.makedirs(args.output_dir) logger.info("Saving model checkpoint to %s", args.output_dir) # Save a trained model, configuration and tokenizer using `save_pretrained()`. # They can then be reloaded using `from_pretrained()` model_to_save = model.module if hasattr( model, 'module') else model # Take care of distributed/parallel training model_to_save.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir) # Good practice: save your training arguments together with the trained model torch.save(args, os.path.join(args.output_dir, 'training_args.bin')) # Load a trained model and vocabulary that you have fine-tuned model = model_class.from_pretrained(args.output_dir, force_download=True, mix_qkv=mix_qkv) tokenizer = tokenizer_class.from_pretrained( args.output_dir, do_lower_case=args.do_lower_case) model.to(args.device) # Evaluation - we can ask to evaluate all the checkpoints (sub-directories) in a directory results = {} if args.do_eval and args.local_rank in [-1, 0]: checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted( glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True))) logging.getLogger("transformers.modeling_utils").setLevel( logging.WARN) # Reduce model loading logs logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: # Reload the model global_step = checkpoint.split( '-')[-1] if len(checkpoints) > 1 else "" if args.mkldnn_eval or args.do_fp32_inference: model = model_class.from_pretrained(checkpoint, force_download=True) model.to(args.device) # Evaluate result, _ = evaluate(args, model, tokenizer, prefix=global_step) result = dict( (k + ('_{}'.format(global_step) if global_step else ''), v) for k, v in result.items()) results.update(result) if args.do_ilit_tune: def eval_func_for_ilit(model): result, _ = evaluate(args, model, tokenizer) for key in sorted(result.keys()): logger.info(" %s = %s", key, str(result[key])) bert_task_acc_keys = [ 'best_f1', 'f1', 'mcc', 'spearmanr', 'acc' ] for key in bert_task_acc_keys: if key in result.keys(): logger.info("Finally Eval {}:{}".format( key, result[key])) acc = result[key] break return acc model = model_class.from_pretrained(checkpoint, force_download=True, mix_qkv=True) model.to(args.device) dataset = load_and_cache_examples(args, tokenizer, evaluate=True, output_examples=False) args.eval_batch_size = args.per_gpu_eval_batch_size * max( 1, args.n_gpu) eval_sampler = SequentialSampler(dataset) eval_dataloader = DataLoader(dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) test_dataloader = Bert_DataLoader(eval_dataloader, args.model_type, args.device) import ilit tuner = ilit.Tuner("./conf.yaml") tuner.tune(model, test_dataloader, eval_func=eval_func_for_ilit) exit(0) if args.do_calibration: model = model_class.from_pretrained(checkpoint, force_download=True, mix_qkv=True) model.to(args.device) model.qconfig = default_per_channel_qconfig propagate_qconfig_(model) add_observer_(model) # Evaluate evaluate(args, model, tokenizer, prefix=global_step, calibration=True) convert(model, inplace=True) quantized_model_path = "squad" + str( global_step) + "_quantized_model" if not os.path.exists(quantized_model_path): os.makedirs(quantized_model_path) model.save_pretrained(quantized_model_path) result, _ = evaluate(args, model, tokenizer, prefix=global_step) result = dict( (k + ('_{}'.format(global_step) if global_step else ''), v) for k, v in result.items()) results.update(result) if args.do_int8_inference: model = model_class.from_pretrained(checkpoint, force_download=True, mix_qkv=True) model.to(args.device) model.qconfig = default_per_channel_qconfig propagate_qconfig_(model) add_observer_(model) convert(model, inplace=True) quantized_model_path = "squad" + str( global_step) + "_quantized_model" if not os.path.exists(quantized_model_path): logger.info("Please run calibration first!") return model_bin_file = os.path.join(quantized_model_path, "pytorch_model.bin") state_dict = torch.load(model_bin_file) model.load_state_dict(state_dict) print(model) with torch.autograd.profiler.profile() as prof: result, _ = evaluate(args, model, tokenizer, prefix=global_step) print(prof.key_averages().table(sort_by="cpu_time_total")) result = dict( (k + ('_{}'.format(global_step) if global_step else ''), v) for k, v in result.items()) results.update(result) logger.info("Results: {}".format(results)) return results
def main(): global args, best_acc1 args = parser.parse_args() print('args:', args) args.distributed = args.world_size > 1 if args.distributed: dist.init_process_group(backend=args.dist_backend, init_method=args.dist_url, world_size=args.world_size) # Val data loading valdir = os.path.join(args.data, 'val') normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) val_dataset = datasets.ImageFolder( valdir, transforms.Compose([ transforms.Resize(args.input_dim + 32), transforms.CenterCrop(args.input_dim), transforms.ToTensor(), normalize, ])) val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers, pin_memory=True) num_classes = len(val_dataset.classes) print('Total classes: ', num_classes) # create model print("=> creating model '{}'".format(args.arch)) if args.arch == 'peleenet': model = PeleeNet(num_classes=num_classes) else: print( "=> unsupported model '{}'. creating PeleeNet by default.".format( args.arch)) model = PeleeNet(num_classes=num_classes) if args.distributed: model = torch.nn.parallel.DistributedDataParallel(model) else: # DataParallel will divide model = torch.nn.DataParallel(model) # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.SGD(model.parameters(), args.lr, momentum=args.momentum, weight_decay=args.weight_decay) # optionally resume from a checkpoint if args.resume: if os.path.isfile(args.resume): print("=> loading checkpoint '{}'".format(args.resume)) checkpoint = torch.load(args.resume) args.start_epoch = checkpoint['epoch'] best_acc1 = checkpoint['best_acc1'] model.load_state_dict(checkpoint['state_dict']) optimizer.load_state_dict(checkpoint['optimizer']) print("=> loaded checkpoint '{}' (epoch {})".format( args.resume, checkpoint['epoch'])) else: print("=> no checkpoint found at '{}'".format(args.resume)) elif args.pretrained: if os.path.isfile(args.weights): checkpoint = torch.load(args.weights, map_location=torch.device('cpu')) model.load_state_dict(checkpoint['state_dict']) print("=> loaded checkpoint '{}' (epoch {}, acc@1 {})".format( args.pretrained, checkpoint['epoch'], checkpoint['best_acc1'])) else: print("=> no checkpoint found at '{}'".format(args.resume)) if args.evaluate: validate(val_loader, model, criterion) return if args.tune: model.eval() model.module.fuse_model() import ilit tuner = ilit.Tuner("./conf.yaml") q_model = tuner.tune(model) exit(0) # Training data loading traindir = os.path.join(args.data, 'train') train_dataset = datasets.ImageFolder( traindir, transforms.Compose([ transforms.RandomResizedCrop(args.input_dim), transforms.RandomHorizontalFlip(), transforms.ToTensor(), normalize, ])) if args.distributed: train_sampler = torch.utils.data.distributed.DistributedSampler( train_dataset) else: train_sampler = None train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None), num_workers=args.workers, pin_memory=True, sampler=train_sampler) for epoch in range(args.start_epoch, args.epochs): if args.distributed: train_sampler.set_epoch(epoch) # train for one epoch train(train_loader, model, criterion, optimizer, epoch) # evaluate on validation set acc1 = validate(val_loader, model, criterion) # remember best Acc@1 and save checkpoint is_best = acc1 > best_acc1 best_acc1 = max(acc1, best_acc1) save_checkpoint( { 'epoch': epoch + 1, 'arch': args.arch, 'state_dict': model.state_dict(), 'best_acc1': best_acc1, 'optimizer': optimizer.state_dict(), }, is_best)
def main(): parser = argparse.ArgumentParser() ## Required parameters parser.add_argument( "--data_dir", default=None, type=str, required=True, help= "The input data dir. Should contain the .tsv files (or other data files) for the task." ) parser.add_argument("--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys())) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list;" + ", ".join(ALL_MODELS)) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys())) parser.add_argument( "--output_dir", default=None, type=str, required=True, help= "The output directory where the model predictions and checkpoints will be written." ) ## Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name") parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name") parser.add_argument( "--cache_dir", default="", type=str, help= "Where do you want to store the pre-trained models downloaded from s3") parser.add_argument( "--max_seq_length", default=128, type=int, help= "The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.") parser.add_argument("--do_train", action='store_true', help="Whether to run training.") parser.add_argument("--do_eval", action='store_true', help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action='store_true', help="Rul evaluation during training at each logging step.") parser.add_argument( "--do_lower_case", action='store_true', help="Set this flag if you are using an uncased model.") parser.add_argument("--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.") parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.") parser.add_argument( '--gradient_accumulation_steps', type=int, default=1, help= "Number of updates steps to accumulate before performing a backward/update pass." ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight deay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument("--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.") parser.add_argument( "--max_steps", default=-1, type=int, help= "If > 0: set total number of training steps to perform. Override num_train_epochs." ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument('--logging_steps', type=int, default=50, help="Log every X updates steps.") parser.add_argument('--save_steps', type=int, default=50, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action='store_true', help= "Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number" ) parser.add_argument("--no_cuda", action='store_true', help="Avoid using CUDA when available") parser.add_argument("--mkldnn_eval", action='store_true', help="evaluation with MKLDNN") parser.add_argument("--mkldnn_train", action='store_true', help="training with MKLDNN") parser.add_argument('--overwrite_output_dir', action='store_true', help="Overwrite the content of the output directory") parser.add_argument( '--overwrite_cache', action='store_true', help="Overwrite the cached training and evaluation sets") parser.add_argument('--seed', type=int, default=42, help="random seed for initialization") parser.add_argument( '--fp16', action='store_true', help= "Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit" ) parser.add_argument( '--fp16_opt_level', type=str, default='O1', help= "For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html") parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument('--server_ip', type=str, default='', help="For distant debugging.") parser.add_argument('--server_port', type=str, default='', help="For distant debugging.") parser.add_argument("--do_fp32_inference", action='store_true', help="Whether to run fp32 inference.") parser.add_argument("--do_calibration", action='store_true', help="Whether to do calibration.") parser.add_argument("--do_int8_inference", action='store_true', help="Whether to run int8 inference.") parser.add_argument("--do_bf16", action='store_true', help="run bf16 evaluation / training.") parser.add_argument("--tune", action='store_true', help="run ilit to tune int8 acc.") parser.add_argument("--warmup", type=int, default=2, help="warmup for performance") args = parser.parse_args() if os.path.exists(args.output_dir) and os.listdir( args.output_dir ) and args.do_train and not args.overwrite_output_dir: raise ValueError( "Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome." .format(args.output_dir)) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend='nccl') args.n_gpu = 1 args.device = device # Setup logging logging.basicConfig( format='%(asctime)s - %(levelname)s - %(name)s - %(message)s', datefmt='%m/%d/%Y %H:%M:%S', level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN) logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) processor = processors[args.task_name]() args.output_mode = output_modes[args.task_name] label_list = processor.get_labels() num_labels = len(label_list) mix_qkv = False if args.do_calibration or args.do_int8_inference or args.tune: mix_qkv = True # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier( ) # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool('.ckpt' in args.model_name_or_path), config=config, mix_qkv=mix_qkv, bf16=args.do_bf16, mkldnn_train=args.mkldnn_train, cache_dir=args.cache_dir if args.cache_dir else None) if args.local_rank == 0: torch.distributed.barrier( ) # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Training if args.do_train: train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, evaluate=False) global_step, tr_loss = train(args, train_dataset, model, tokenizer) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained() if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0): # Create output directory if needed if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]: os.makedirs(args.output_dir) logger.info("Saving model checkpoint to %s", args.output_dir) # Save a trained model, configuration and tokenizer using `save_pretrained()`. # They can then be reloaded using `from_pretrained()` model_to_save = model.module if hasattr( model, 'module') else model # Take care of distributed/parallel training model_to_save.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir) # Good practice: save your training arguments together with the trained model torch.save(args, os.path.join(args.output_dir, 'training_args.bin')) # Load a trained model and vocabulary that you have fine-tuned model = model_class.from_pretrained(args.output_dir) tokenizer = tokenizer_class.from_pretrained(args.output_dir) model.to(args.device) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained( args.output_dir, do_lower_case=args.do_lower_case) checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted( glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True))) logging.getLogger("transformers.modeling_utils").setLevel( logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split( '-')[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split( '/')[-1] if checkpoint.find('checkpoint') != -1 else "" logger.info("Evaluate:" + args.task_name) if args.mkldnn_eval or args.do_fp32_inference or args.do_bf16: model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, prefix=prefix) result = dict((k + '_{}'.format(global_step), v) for k, v in result.items()) results.update(result) if args.tune: def eval_func_for_ilit(model): result, perf = evaluate(args, model, tokenizer, prefix=prefix) bert_task_acc_keys = [ 'acc_and_f1', 'f1', 'mcc', 'spearmanr', 'acc' ] for key in bert_task_acc_keys: if key in result.keys(): logger.info("Finally Eval {}:{}".format( key, result[key])) acc = result[key] break return acc model = model_class.from_pretrained(checkpoint, mix_qkv=True) model.to(args.device) eval_task_names = ( "mnli", "mnli-mm") if args.task_name == "mnli" else ( args.task_name, ) for eval_task in eval_task_names: eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, evaluate=True) args.eval_batch_size = args.per_gpu_eval_batch_size * max( 1, args.n_gpu) # multi-gpu eval if args.n_gpu > 1: model = torch.nn.DataParallel(model) if args.mkldnn_eval: from torch.utils import mkldnn as mkldnn_utils model = mkldnn_utils.to_mkldnn(model) print(model) import ilit tuner = ilit.Tuner("./conf.yaml") if eval_task != "squad": eval_task = 'classifier' eval_dataset = tuner.dataset('bert', dataset=eval_dataset, task=eval_task) test_dataloader = tuner.dataloader( eval_dataset, batch_size=args.eval_batch_size) tuner.tune(model, test_dataloader, eval_func=eval_func_for_ilit) exit(0) if args.do_calibration: model = model_class.from_pretrained(checkpoint, mix_qkv=True) model.to(args.device) model.qconfig = default_per_channel_qconfig fallback_layers = {} if args.model_name_or_path == "bert-base-uncased" and args.task_name == "mrpc": fallback_layers = {"bert.encoder.layer.9.output.dense."} propagate_qconfig_(model) fallback_layer(model, layer_name="", exculde_layers=fallback_layers) add_observer_(model) result, _ = evaluate(args, model, tokenizer, prefix=global_step, calibration=True) convert(model, inplace=True) quantized_model_path = args.task_name + "_quantized_model" if not os.path.exists(quantized_model_path): os.makedirs(quantized_model_path) model.save_pretrained(quantized_model_path) print(model) result, _ = evaluate(args, model, tokenizer, prefix=prefix) if args.do_int8_inference: model = model_class.from_pretrained(checkpoint, mix_qkv=True) model.to(args.device) model.qconfig = default_per_channel_qconfig fallback_layers = {} if args.model_name_or_path == "bert-base-uncased" and args.task_name == "mrpc": fallback_layers = {"bert.encoder.layer.9.output.dense."} propagate_qconfig_(model) fallback_layer(model, layer_name="", exculde_layers=fallback_layers) add_observer_(model) convert(model, inplace=True) quantized_model_path = args.task_name + "_quantized_model" if not os.path.exists(quantized_model_path): logger.error( "please do calibrantion befor run int8 inference") return prepare(model, inplace=True) convert(model, inplace=True) model_bin_file = os.path.join(quantized_model_path, "pytorch_model.bin") state_dict = torch.load(model_bin_file) model.load_state_dict(state_dict) result, _ = evaluate(args, model, tokenizer, prefix=prefix) return results
arg_parser.add_argument('-w', "--warmup_iter", help='For benchmark measurement only.', dest='warmup_iter', default=200, type=int) arg_parser.add_argument('--config', type=str, default='') arg_parser.add_argument('--output_model', type=str, default='') arg_parser.add_argument('--tune', action='store_true', default=False) arg_parser.add_argument('--benchmark', dest='benchmark', action='store_true', help='run benchmark') args = arg_parser.parse_args() infer = model_infer(args) if args.tune: at = ilit.Tuner(args.config) q_dataloader = at.dataloader(infer, args.batch_size) output_graph = at.tune(infer.get_graph(), q_dataloader=q_dataloader, eval_func=infer.accuracy_check) try: write_graph(output_graph.as_graph_def(), args.output_model) except Exception as e: logging.getLogger().info("Failed to save model due to {}".format( str(e))) else: infer.run()