This repository has been archived by the owner on Jul 24, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 1
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge branch 'main' into egor/ipex_config
- Loading branch information
Showing
9 changed files
with
496 additions
and
380 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,172 +1,46 @@ | ||
| import time | ||
| from typing import List | ||
|
|
||
| import numpy as np | ||
| import torch | ||
| import torch.nn as nn | ||
| import torch.optim as optim | ||
| from torch.utils.data import DataLoader, Dataset | ||
|
|
||
| from dl_bench.utils import Backend, Benchmark, TimerManager | ||
|
|
||
| from dl_bench.mlp import get_macs | ||
|
|
||
|
|
||
| def get_time(): | ||
| return time.perf_counter() | ||
|
|
||
|
|
||
| # PARAMS | ||
| class RandomInfDataset(Dataset): | ||
| def __init__(self, n, in_shape): | ||
| super().__init__() | ||
|
|
||
| self.values = np.random.randn(n, *in_shape).astype(np.float32) | ||
|
|
||
| def __len__(self): | ||
| return len(self.values) | ||
|
|
||
| def __getitem__(self, index): | ||
| return self.values[index] | ||
|
|
||
|
|
||
| def get_inf_loaders(n, in_shape, batch_size, device: str): | ||
| # This speeds up data copy for cuda devices | ||
| pin_memory = device == "cuda" | ||
|
|
||
| ds = RandomInfDataset(n, in_shape) | ||
| train_loader = DataLoader( | ||
| ds, batch_size=batch_size, shuffle=True, num_workers=4, pin_memory=pin_memory | ||
| ) | ||
| test_loader = DataLoader( | ||
| ds, batch_size=batch_size, shuffle=False, num_workers=4, pin_memory=pin_memory | ||
| ) | ||
| return train_loader, test_loader | ||
| from dl_bench.utils import Benchmark, RandomInfDataset | ||
|
|
||
|
|
||
| def get_cnn(name): | ||
| from torchvision.models import vgg16, resnet18, resnet50, resnext50_32x4d, resnext101_32x8d, densenet121, efficientnet_v2_m, mobilenet_v3_large | ||
| from torchvision.models import ( | ||
| vgg16, | ||
| resnet18, | ||
| resnet50, | ||
| resnext50_32x4d, | ||
| resnext101_32x8d, | ||
| densenet121, | ||
| efficientnet_v2_m, | ||
| mobilenet_v3_large, | ||
| ) | ||
|
|
||
| name2model = { | ||
| 'vgg16': vgg16, | ||
| 'resnet18': resnet18, | ||
| 'resnet50': resnet50, | ||
| 'resnext50': resnext50_32x4d, | ||
| 'resnext101': resnext101_32x8d, | ||
| 'densenet121': densenet121, | ||
| 'efficientnet_v2m': efficientnet_v2_m, | ||
| 'mobilenet_v3l': mobilenet_v3_large, | ||
| "vgg16": vgg16, | ||
| "resnet18": resnet18, | ||
| "resnet50": resnet50, | ||
| "resnext50": resnext50_32x4d, | ||
| "resnext101": resnext101_32x8d, | ||
| "densenet121": densenet121, | ||
| "efficientnet_v2m": efficientnet_v2_m, | ||
| "mobilenet_v3_large": mobilenet_v3_large, | ||
| } | ||
| if name in name2model: | ||
| return name2model[name]() | ||
| else: | ||
| raise ValueError(f"Unknown name {name}") | ||
|
|
||
|
|
||
| def build_mlp( | ||
| n_chans_in: int, | ||
| n_chans_out: int, | ||
| struct: List[int], | ||
| norm_layer=None, | ||
| activ_layer=nn.ReLU, | ||
| inplace=None, | ||
| dropout: float = None, | ||
| ): | ||
| params = {} if inplace is None else {"inplace": inplace} | ||
| bias = True if dropout is None else False | ||
|
|
||
| layers = [] | ||
| in_dim = n_chans_in | ||
| for hidden_dim in struct: | ||
| layers.append(torch.nn.Linear(in_dim, hidden_dim, bias=bias)) | ||
| if norm_layer is not None: | ||
| layers.append(norm_layer(hidden_dim)) | ||
| if activ_layer is not None: | ||
| layers.append(activ_layer(**params)) | ||
| if dropout is not None: | ||
| layers.append(torch.nn.Dropout(dropout, **params)) | ||
| in_dim = hidden_dim | ||
|
|
||
| layers.append(torch.nn.Linear(in_dim, n_chans_out, bias=True)) | ||
|
|
||
| return nn.Sequential(*layers) | ||
|
|
||
|
|
||
| def get_mlp(n_chans_in, n_chans_out, name): | ||
| params = name2params[name] | ||
|
|
||
| # net = nn.Sequential(nn.Flatten(), build_mlp(n_chans_in, **params)) | ||
| net = build_mlp(n_chans_in, **params, n_chans_out=n_chans_out) | ||
| return net | ||
|
|
||
|
|
||
| IN_SHAPE = (3, 224, 224) | ||
|
|
||
| class CnnBenchmark(Benchmark): | ||
| def run(self, backend: Backend, params): | ||
| tm = TimerManager() | ||
|
|
||
| # PARAMS | ||
| name = params.get("name", "resnet50") | ||
| def __init__(self, params) -> None: | ||
| batch_size = int(params.get("batch_size", 1024)) | ||
|
|
||
| # Do early stopping once we hit min_batches & min_seconds to accelerate measurement | ||
| in_shape = (3, 224, 224) | ||
| min_batches = 10 | ||
| min_seconds = 10 | ||
| DATASET_SIZE = max(10_240, batch_size * min_batches) | ||
| dataset = RandomInfDataset(DATASET_SIZE, in_shape) | ||
|
|
||
| trainloader, testloader = get_inf_loaders( | ||
| DATASET_SIZE, | ||
| in_shape=IN_SHAPE, | ||
| batch_size=batch_size, | ||
| device=backend.device_name, | ||
| ) | ||
| name = params.get("name", "resnet50") | ||
| net = get_cnn(name=name) | ||
| # flops_per_sample = get_macs(net, IN_SHAPE, backend) * 2 | ||
| flops_per_sample = 2 * get_macs(net, IN_SHAPE, backend) | ||
|
|
||
| sample = backend.to_device(torch.rand(batch_size, *IN_SHAPE)) | ||
| net = backend.prepare_eval_model(net, sample_input=sample) | ||
| print("Warmup started") | ||
| with torch.no_grad(): | ||
| net.eval() | ||
| with tm.timeit("warmup_s"): | ||
| net(sample) | ||
| print("Warmup done") | ||
|
|
||
| correct = 0 | ||
| total = 0 | ||
| n_items = 0 | ||
|
|
||
| net.eval() | ||
| with torch.no_grad(): | ||
| start = time.perf_counter() | ||
| with tm.timeit("duration_s"): | ||
| for x in testloader: | ||
| x = backend.to_device(x) | ||
| if backend.dtype == torch.float32: | ||
| output = net(x) | ||
| assert output.dtype is backend.dtype, f"{output.dtype}!={backend.dtype}" | ||
| _, predicted = torch.max(output.data, 1) | ||
|
|
||
| else: | ||
| with torch.autocast(device_type=backend.device_name, dtype=backend.dtype): | ||
| output = net(x) | ||
| assert output.dtype is backend.dtype, f"{output.dtype}!={backend.dtype}" | ||
| _, predicted = torch.max(output.data, 1) | ||
|
|
||
| n_items += len(x) | ||
|
|
||
| # early stopping | ||
| if (time.perf_counter() - start) > min_seconds and n_items > batch_size * min_batches: | ||
| break | ||
|
|
||
| print(f"{n_items} were processed in {tm.name2time['duration_s']}s") | ||
|
|
||
| results = tm.get_results() | ||
| results["samples_per_s"] = n_items / results["duration_s"] | ||
| results["flops_per_sample"] = flops_per_sample | ||
|
|
||
| return results | ||
|
|
||
| # print(f'Accuracy of the network on the 10000 test images: {100 * correct // total} %') | ||
| super().__init__( | ||
| net=net, in_shape=in_shape, dataset=dataset, batch_size=batch_size | ||
| ) |
Oops, something went wrong.