-
Notifications
You must be signed in to change notification settings - Fork 364
chore: cherry-pick FP8 #2892
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
chore: cherry-pick FP8 #2892
Changes from all commits
Commits
Show all changes
3 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or 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 hidden or 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 |
|---|---|---|
| @@ -1,3 +1,3 @@ | ||
| __version__: "2.4.0.dev0" | ||
| __cuda_version__: "12.1" | ||
| __tensorrt_version__: "10.0.1" | ||
| __tensorrt_version__: "10.0.1" |
This file contains hidden or 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 hidden or 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 hidden or 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 |
|---|---|---|
| @@ -0,0 +1,251 @@ | ||
| """ | ||
| .. _vgg16_fp8_ptq: | ||
|
|
||
| Torch Compile VGG16 with FP8 and PTQ | ||
| ====================================================== | ||
|
|
||
| This script is intended as a sample of the Torch-TensorRT workflow with `torch.compile` on a VGG16 model with FP8 and PTQ. | ||
| """ | ||
|
|
||
| # %% | ||
| # Imports and Model Definition | ||
| # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | ||
|
|
||
| import argparse | ||
|
|
||
| import modelopt.torch.quantization as mtq | ||
| import torch | ||
| import torch.nn as nn | ||
| import torch.nn.functional as F | ||
| import torch_tensorrt as torchtrt | ||
| import torchvision.datasets as datasets | ||
| import torchvision.transforms as transforms | ||
| from modelopt.torch.quantization.utils import export_torch_mode | ||
|
|
||
|
|
||
| class VGG(nn.Module): | ||
| def __init__(self, layer_spec, num_classes=1000, init_weights=False): | ||
| super(VGG, self).__init__() | ||
|
|
||
| layers = [] | ||
| in_channels = 3 | ||
| for l in layer_spec: | ||
| if l == "pool": | ||
| layers.append(nn.MaxPool2d(kernel_size=2, stride=2)) | ||
| else: | ||
| layers += [ | ||
| nn.Conv2d(in_channels, l, kernel_size=3, padding=1), | ||
| nn.BatchNorm2d(l), | ||
| nn.ReLU(), | ||
| ] | ||
| in_channels = l | ||
|
|
||
| self.features = nn.Sequential(*layers) | ||
| self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) | ||
| self.classifier = nn.Sequential( | ||
| nn.Linear(512 * 1 * 1, 4096), | ||
| nn.ReLU(), | ||
| nn.Dropout(), | ||
| nn.Linear(4096, 4096), | ||
| nn.ReLU(), | ||
| nn.Dropout(), | ||
| nn.Linear(4096, num_classes), | ||
| ) | ||
| if init_weights: | ||
| self._initialize_weights() | ||
|
|
||
| def _initialize_weights(self): | ||
| for m in self.modules(): | ||
| if isinstance(m, nn.Conv2d): | ||
| nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu") | ||
| if m.bias is not None: | ||
| nn.init.constant_(m.bias, 0) | ||
| elif isinstance(m, nn.BatchNorm2d): | ||
| nn.init.constant_(m.weight, 1) | ||
| nn.init.constant_(m.bias, 0) | ||
| elif isinstance(m, nn.Linear): | ||
| nn.init.normal_(m.weight, 0, 0.01) | ||
| nn.init.constant_(m.bias, 0) | ||
|
|
||
| def forward(self, x): | ||
| x = self.features(x) | ||
| x = self.avgpool(x) | ||
| x = torch.flatten(x, 1) | ||
| x = self.classifier(x) | ||
| return x | ||
|
|
||
|
|
||
| def vgg16(num_classes=1000, init_weights=False): | ||
| vgg16_cfg = [ | ||
| 64, | ||
| 64, | ||
| "pool", | ||
| 128, | ||
| 128, | ||
| "pool", | ||
| 256, | ||
| 256, | ||
| 256, | ||
| "pool", | ||
| 512, | ||
| 512, | ||
| 512, | ||
| "pool", | ||
| 512, | ||
| 512, | ||
| 512, | ||
| "pool", | ||
| ] | ||
| return VGG(vgg16_cfg, num_classes, init_weights) | ||
|
|
||
|
|
||
| PARSER = argparse.ArgumentParser( | ||
| description="Load pre-trained VGG model and then tune with FP8 and PTQ" | ||
| ) | ||
| PARSER.add_argument( | ||
| "--ckpt", type=str, required=True, help="Path to the pre-trained checkpoint" | ||
| ) | ||
| PARSER.add_argument( | ||
| "--batch-size", | ||
| default=128, | ||
| type=int, | ||
| help="Batch size for tuning the model with PTQ and FP8", | ||
| ) | ||
|
|
||
| args = PARSER.parse_args() | ||
|
|
||
| model = vgg16(num_classes=10, init_weights=False) | ||
| model = model.cuda() | ||
|
|
||
| # %% | ||
| # Load the pre-trained model weights | ||
| # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | ||
|
|
||
| ckpt = torch.load(args.ckpt) | ||
| weights = ckpt["model_state_dict"] | ||
|
|
||
| if torch.cuda.device_count() > 1: | ||
| from collections import OrderedDict | ||
|
|
||
| new_state_dict = OrderedDict() | ||
| for k, v in weights.items(): | ||
| name = k[7:] # remove `module.` | ||
| new_state_dict[name] = v | ||
| weights = new_state_dict | ||
|
|
||
| model.load_state_dict(weights) | ||
| # Don't forget to set the model to evaluation mode! | ||
| model.eval() | ||
|
|
||
| # %% | ||
| # Load training dataset and define loss function for PTQ | ||
| # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | ||
|
|
||
| training_dataset = datasets.CIFAR10( | ||
| root="./data", | ||
| train=True, | ||
| download=True, | ||
| transform=transforms.Compose( | ||
| [ | ||
| transforms.RandomCrop(32, padding=4), | ||
| transforms.RandomHorizontalFlip(), | ||
| transforms.ToTensor(), | ||
| transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)), | ||
| ] | ||
| ), | ||
| ) | ||
| training_dataloader = torch.utils.data.DataLoader( | ||
| training_dataset, batch_size=args.batch_size, shuffle=True, num_workers=2 | ||
| ) | ||
|
|
||
| data = iter(training_dataloader) | ||
| images, _ = next(data) | ||
|
|
||
| crit = nn.CrossEntropyLoss() | ||
|
|
||
| # %% | ||
| # Define Calibration Loop for quantization | ||
| # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | ||
|
|
||
|
|
||
| def calibrate_loop(model): | ||
| # calibrate over the training dataset | ||
| total = 0 | ||
| correct = 0 | ||
| loss = 0.0 | ||
| for data, labels in training_dataloader: | ||
| data, labels = data.cuda(), labels.cuda(non_blocking=True) | ||
| out = model(data) | ||
| loss += crit(out, labels) | ||
| preds = torch.max(out, 1)[1] | ||
| total += labels.size(0) | ||
| correct += (preds == labels).sum().item() | ||
|
|
||
| print("PTQ Loss: {:.5f} Acc: {:.2f}%".format(loss / total, 100 * correct / total)) | ||
|
|
||
|
|
||
| # %% | ||
| # Tune the pre-trained model with FP8 and PTQ | ||
| # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | ||
|
|
||
| quant_cfg = mtq.FP8_DEFAULT_CFG | ||
| # PTQ with in-place replacement to quantized modules | ||
| mtq.quantize(model, quant_cfg, forward_loop=calibrate_loop) | ||
| # model has FP8 qdq nodes at this point | ||
|
|
||
| # %% | ||
| # Inference | ||
| # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | ||
|
|
||
| # Load the testing dataset | ||
| testing_dataset = datasets.CIFAR10( | ||
| root="./data", | ||
| train=False, | ||
| download=True, | ||
| transform=transforms.Compose( | ||
| [ | ||
| transforms.ToTensor(), | ||
| transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)), | ||
| ] | ||
| ), | ||
| ) | ||
|
|
||
| testing_dataloader = torch.utils.data.DataLoader( | ||
| testing_dataset, batch_size=args.batch_size, shuffle=False, num_workers=2 | ||
| ) | ||
|
|
||
| with torch.no_grad(): | ||
| with export_torch_mode(): | ||
| # Compile the model with Torch-TensorRT Dynamo backend | ||
| input_tensor = images.cuda() | ||
| exp_program = torch.export.export(model, (input_tensor,)) | ||
| trt_model = torchtrt.dynamo.compile( | ||
| exp_program, | ||
| inputs=[input_tensor], | ||
| enabled_precisions={torch.float8_e4m3fn}, | ||
| min_block_size=1, | ||
| debug=False, | ||
| ) | ||
|
|
||
| # Inference compiled Torch-TensorRT model over the testing dataset | ||
| total = 0 | ||
| correct = 0 | ||
| loss = 0.0 | ||
| class_probs = [] | ||
| class_preds = [] | ||
| model.eval() | ||
| for data, labels in testing_dataloader: | ||
| data, labels = data.cuda(), labels.cuda(non_blocking=True) | ||
| out = model(data) | ||
| loss += crit(out, labels) | ||
| preds = torch.max(out, 1)[1] | ||
| class_probs.append([F.softmax(i, dim=0) for i in out]) | ||
| class_preds.append(preds) | ||
| total += labels.size(0) | ||
| correct += (preds == labels).sum().item() | ||
|
|
||
| test_probs = torch.cat([torch.stack(batch) for batch in class_probs]) | ||
| test_preds = torch.cat(class_preds) | ||
| test_loss = loss / total | ||
| test_acc = correct / total | ||
| print("Test Loss: {:.5f} Test Acc: {:.2f}%".format(test_loss, 100 * test_acc)) |
This file contains hidden or 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 hidden or 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
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.