Skip to content

Commit 9255eb3

Browse files
asfiyab-nvidiaasfiyab-nvidia
authored
TensorRT 10.11 release updates (#4455)
* TensorRT 10.11 release updates Signed-off-by: Asfiya Baig <[email protected]> * Update changelog date Signed-off-by: Asfiya Baig <[email protected]> * Update ONNX parser Signed-off-by: Asfiya Baig <[email protected]> * add shouldCompileKernel Signed-off-by: Asfiya Baig <[email protected]> * changelog updates Signed-off-by: Asfiya Baig <[email protected]> * Update changelog plugin Signed-off-by: Asfiya Baig <[email protected]> --------- Signed-off-by: Asfiya Baig <[email protected]>
1 parent b71ada5 commit 9255eb3

File tree

126 files changed

+6536
-1223
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

126 files changed

+6536
-1223
lines changed

‎CHANGELOG.md

Lines changed: 17 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,22 @@
11
# TensorRT OSS Release Changelog
22

3+
## 10.11.0 GA - 2025-5-21
4+
5+
Key Features and Updates:
6+
7+
- Plugin changes
8+
- Migrated `IPluginV2`-descendent version 1 of `modulatedDeformConvPlugin`, to version 2, which implements `IPluginV3`.
9+
- Migrated `IPluginV2`-descendent version 1 of `DisentangledAttention_TRT`, to version 2, which implements `IPluginV3`.
10+
- Migrated `IPluginV2`-descendent version 1 of `MultiscaleDeformableAttnPlugin_TRT`, to version 2, which implements `IPluginV3`.
11+
- Note: The newer versions preserve the attributes and I/O of the corresponding older plugin version. The older plugin versions are deprecated and will be removed in a future release.
12+
- Demo changes
13+
- demoDiffusion
14+
- Added support for Stable Diffusion 3.5-medium and 3.5-large pipelines in BF16 and FP16 precisions.
15+
- Parser changes
16+
- Added `kENABLE_UINT8_AND_ASYMMETRIC_QUANTIZATION_DLA` parser flag to enable UINT8 asymmetric quantization on engines targeting DLA.
17+
- Removed restriction that inputs to `RandomNormalLike` and `RandomUniformLike` must be tensors.
18+
- Clarified limitations of scan outputs for `Loop` nodes.
19+
320
## 10.10.0 GA - 2025-4-28
421

522
Key Features and Updates:

‎CMakeLists.txt

Lines changed: 15 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,7 @@
1818
cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
1919
include(cmake/modules/set_ifndef.cmake)
2020
include(cmake/modules/find_library_create_target.cmake)
21+
list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake/modules)
2122

2223
set_ifndef(TRT_LIB_DIR ${CMAKE_BINARY_DIR})
2324
set_ifndef(TRT_OUT_DIR ${CMAKE_BINARY_DIR})
@@ -47,10 +48,10 @@ else()
4748
set(STATIC_LIB_EXT "a")
4849
endif()
4950

50-
file(STRINGS "${CMAKE_CURRENT_SOURCE_DIR}/include/NvInferVersion.h" VERSION_STRINGS REGEX "#define NV_TENSORRT_.*")
51+
file(STRINGS "${CMAKE_CURRENT_SOURCE_DIR}/include/NvInferVersion.h" VERSION_STRINGS REGEX "#define TRT_.*_ENTERPRISE")
5152

5253
foreach(TYPE MAJOR MINOR PATCH BUILD)
53-
string(REGEX MATCH "NV_TENSORRT_${TYPE} [0-9]+" TRT_TYPE_STRING ${VERSION_STRINGS})
54+
string(REGEX MATCH "TRT_${TYPE}_ENTERPRISE [0-9]+" TRT_TYPE_STRING ${VERSION_STRINGS})
5455
string(REGEX MATCH "[0-9]+" TRT_${TYPE} ${TRT_TYPE_STRING})
5556
endforeach(TYPE)
5657

@@ -143,20 +144,25 @@ if(BUILD_PARSERS)
143144
configure_protobuf(${PROTOBUF_VERSION})
144145
endif()
145146

147+
# Define library names
148+
set(TRT_NVINFER_NAME "nvinfer")
149+
set(TRT_ONNXPARSER_NAME "nvonnxparser")
150+
146151
# Windows library names have major version appended.
147152
if (MSVC)
148-
set(nvinfer_lib_name "nvinfer_${TRT_SOVERSION}")
153+
set(nvinfer_lib_name "${TRT_NVINFER_NAME}_${TRT_SOVERSION}${TRT_LIB_SUFFIX}")
149154
set(nvinfer_plugin_lib_name "nvinfer_plugin_${TRT_SOVERSION}")
150155
set(nvinfer_vc_plugin_lib_name "nvinfer_vc_plugin_${TRT_SOVERSION}")
151-
set(nvonnxparser_lib_name "nvonnxparser_${TRT_SOVERSION}")
156+
set(nvonnxparser_lib_name "${TRT_ONNXPARSER_NAME}_${TRT_SOVERSION}${TRT_LIB_SUFFIX}")
157+
152158
else()
153-
set(nvinfer_lib_name "nvinfer")
159+
set(nvinfer_lib_name ${TRT_NVINFER_NAME})
154160
set(nvinfer_plugin_lib_name "nvinfer_plugin")
155161
set(nvinfer_vc_plugin_lib_name "nvinfer_vc_plugin")
156-
set(nvonnxparser_lib_name "nvonnxparser")
162+
set(nvonnxparser_lib_name ${TRT_ONNXPARSER_NAME})
157163
endif()
158164

159-
find_library_create_target(nvinfer ${nvinfer_lib_name} SHARED ${TRT_LIB_DIR})
165+
find_library_create_target(nvinfer ${nvinfer_lib_name} SHARED "${TRT_LIB_DIR}")
160166

161167
if (DEFINED USE_CUGFX)
162168
find_library(CUDART_LIB cugfx_dll HINTS ${CUDA_TOOLKIT_ROOT_DIR} PATH_SUFFIXES lib lib/x64 lib64)
@@ -217,13 +223,13 @@ endif()
217223
if(BUILD_PLUGINS)
218224
add_subdirectory(plugin)
219225
else()
220-
find_library_create_target(nvinfer_plugin ${nvinfer_plugin_lib_name} SHARED ${TRT_OUT_DIR} ${TRT_LIB_DIR})
226+
find_library_create_target(nvinfer_plugin ${nvinfer_plugin_lib_name} SHARED "${TRT_OUT_DIR}" "${TRT_LIB_DIR}")
221227
endif()
222228

223229
if(BUILD_PARSERS)
224230
add_subdirectory(parsers)
225231
else()
226-
find_library_create_target(nvonnxparser ${nvonnxparser_lib_name} SHARED ${TRT_OUT_DIR} ${TRT_LIB_DIR})
232+
find_library_create_target(nvonnxparser ${nvonnxparser_lib_name} SHARED "${TRT_OUT_DIR}" "${TRT_LIB_DIR}")
227233
endif()
228234

229235
if(BUILD_SAMPLES)

‎README.md

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -32,7 +32,7 @@ To build the TensorRT-OSS components, you will first need the following software
3232

3333
**TensorRT GA build**
3434

35-
- TensorRT v10.10.0.31
35+
- TensorRT v10.11.0.33
3636
- Available from direct download links listed below
3737

3838
**System Packages**
@@ -86,24 +86,24 @@ To build the TensorRT-OSS components, you will first need the following software
8686

8787
Else download and extract the TensorRT GA build from [NVIDIA Developer Zone](https://developer.nvidia.com) with the direct links below:
8888

89-
- [TensorRT 10.10.0.31 for CUDA 11.8, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.10.0/tars/TensorRT-10.10.0.31.Linux.x86_64-gnu.cuda-11.8.tar.gz)
90-
- [TensorRT 10.10.0.31 for CUDA 12.9, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.10.0/tars/TensorRT-10.10.0.31.Linux.x86_64-gnu.cuda-12.9.tar.gz)
91-
- [TensorRT 10.10.0.31 for CUDA 11.8, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.10.0/zip/TensorRT-10.10.0.31.Windows.win10.cuda-11.8.zip)
92-
- [TensorRT 10.10.0.31 for CUDA 12.9, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.10.0/zip/TensorRT-10.10.0.31.Windows.win10.cuda-12.9.zip)
89+
- [TensorRT 10.11.0.33 for CUDA 11.8, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.11.0/tars/TensorRT-10.11.0.33.Linux.x86_64-gnu.cuda-11.8.tar.gz)
90+
- [TensorRT 10.11.0.33 for CUDA 12.9, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.11.0/tars/TensorRT-10.11.0.33.Linux.x86_64-gnu.cuda-12.9.tar.gz)
91+
- [TensorRT 10.11.0.33 for CUDA 11.8, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.11.0/zip/TensorRT-10.11.0.33.Windows.win10.cuda-11.8.zip)
92+
- [TensorRT 10.11.0.33 for CUDA 12.9, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.11.0/zip/TensorRT-10.11.0.33.Windows.win10.cuda-12.9.zip)
9393

9494
**Example: Ubuntu 20.04 on x86-64 with cuda-12.9**
9595

9696
```bash
9797
cd ~/Downloads
98-
tar -xvzf TensorRT-10.10.0.31.Linux.x86_64-gnu.cuda-12.9.tar.gz
99-
export TRT_LIBPATH=`pwd`/TensorRT-10.10.0.31
98+
tar -xvzf TensorRT-10.11.0.33.Linux.x86_64-gnu.cuda-12.9.tar.gz
99+
export TRT_LIBPATH=`pwd`/TensorRT-10.11.0.33
100100
```
101101

102102
**Example: Windows on x86-64 with cuda-12.9**
103103

104104
```powershell
105-
Expand-Archive -Path TensorRT-10.10.0.31.Windows.win10.cuda-12.9.zip
106-
$env:TRT_LIBPATH="$pwd\TensorRT-10.10.0.31\lib"
105+
Expand-Archive -Path TensorRT-10.11.0.33.Windows.win10.cuda-12.9.zip
106+
$env:TRT_LIBPATH="$pwd\TensorRT-10.11.0.33\lib"
107107
```
108108

109109
## Setting Up The Build Environment

‎VERSION

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
10.10.0.31
1+
10.11.0.33

‎cmake/modules/ShouldCompileKernel.cmake

Lines changed: 41 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,41 @@
1+
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2+
# SPDX-License-Identifier: Apache-2.0
3+
#
4+
# Licensed under the Apache License, Version 2.0 (the "License");
5+
# you may not use this file except in compliance with the License.
6+
# You may obtain a copy of the License at
7+
#
8+
# http://www.apache.org/licenses/LICENSE-2.0
9+
#
10+
# Unless required by applicable law or agreed to in writing, software
11+
# distributed under the License is distributed on an "AS IS" BASIS,
12+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13+
# See the License for the specific language governing permissions and
14+
# limitations under the License.
15+
16+
# Certain cubins are binary compatible between different SM versions, so they are reused.
17+
# This function checks if a SM-named file should be compiled based on current SM enablement.
18+
# Specifically, the SM80 files are compiled if either 80, 86, or 89 are enabled.
19+
function(should_compile_kernel SM OUT_VAR)
20+
# If the target SM is any of 80/86/89, we need to check if any of those are enabled in CMAKE_CUDA_ARCHITECTURES.
21+
if((${SM} EQUAL 80) OR (${SM} EQUAL 86) OR (${SM} EQUAL 89))
22+
list(FIND CMAKE_CUDA_ARCHITECTURES 80 SM80_INDEX)
23+
list(FIND CMAKE_CUDA_ARCHITECTURES 86 SM86_INDEX)
24+
list(FIND CMAKE_CUDA_ARCHITECTURES 89 SM89_INDEX)
25+
if((NOT ${SM80_INDEX} EQUAL -1) OR
26+
(NOT ${SM86_INDEX} EQUAL -1) OR
27+
(NOT ${SM89_INDEX} EQUAL -1)
28+
)
29+
set(${OUT_VAR} TRUE PARENT_SCOPE)
30+
else()
31+
set(${OUT_VAR} FALSE PARENT_SCOPE)
32+
endif()
33+
else()
34+
list(FIND CMAKE_CUDA_ARCHITECTURES ${SM} SM_INDEX)
35+
if (NOT ${SM_INDEX} EQUAL -1)
36+
set(${OUT_VAR} TRUE PARENT_SCOPE)
37+
else()
38+
set(${OUT_VAR} FALSE PARENT_SCOPE)
39+
endif()
40+
endif()
41+
endfunction()

‎demo/BERT/README.md

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -73,8 +73,8 @@ The following software version configuration has been tested:
7373
| Software | Version |
7474
| -------- | ------- |
7575
| Python | >=3.8 |
76-
| TensorRT | 10.9 |
77-
| CUDA | 12.8 |
76+
| TensorRT | 10.11 |
77+
| CUDA | 12.9 |
7878

7979
## Setup
8080

‎demo/BERT/builder_varseqlen.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -431,7 +431,8 @@ def build_engine(batch_sizes, workspace_size, sequence_length, config, weights_d
431431
network_creation_flag = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
432432

433433
with trt.Builder(TRT_LOGGER) as builder, builder.create_network(network_creation_flag) as network, builder.create_builder_config() as builder_config:
434-
builder_config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, workspace_size * (1024 * 1024))
434+
if workspace_size is not None:
435+
builder_config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, workspace_size * (1024 * 1024))
435436
builder_config.avg_timing_iterations = 8
436437
if config.use_fp16:
437438
builder_config.set_flag(trt.BuilderFlag.FP16)
@@ -571,8 +572,7 @@ def main():
571572
parser.add_argument(
572573
"-w",
573574
"--workspace-size",
574-
default=2500,
575-
help="Workspace size in MiB for building the BERT engine (default: 2500)",
575+
help="Workspace size in MiB for building the BERT engine (default: unlimited)",
576576
type=int,
577577
)
578578
parser.add_argument(

‎demo/Diffusion/README.md

Lines changed: 15 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@ This demo application ("demoDiffusion") showcases the acceleration of Stable Dif
77
### Clone the TensorRT OSS repository
88

99
```bash
10-
git clone [email protected]:NVIDIA/TensorRT.git -b release/10.9 --single-branch
10+
git clone [email protected]:NVIDIA/TensorRT.git -b release/10.11 --single-branch
1111
cd TensorRT
1212
```
1313

@@ -49,7 +49,7 @@ onnx 1.15.0
4949
onnx-graphsurgeon 0.5.2
5050
onnxruntime 1.16.3
5151
polygraphy 0.49.9
52-
tensorrt 10.9.0.34
52+
tensorrt 10.11.0.33
5353
tokenizers 0.13.3
5454
torch 2.2.0
5555
transformers 4.42.2
@@ -199,19 +199,27 @@ Even faster image generation than LCM, producing coherent images in just 1 step.
199199
python3 demo_txt2img_xl.py "Einstein" --version xl-turbo --onnx-dir onnx-sdxl-turbo --engine-dir engine-sdxl-turbo --denoising-steps 1 --scheduler EulerA --guidance-scale 0.0 --width 512 --height 512
200200
```
201201

202-
### Generate an image guided by a text prompt using Stable Diffusion 3
202+
### Generate an image guided by a text prompt using Stable Diffusion 3 and its variants
203203

204-
Run the command below to generate an image using Stable Diffusion 3
204+
Run the command below to generate an image using Stable Diffusion 3 and Stable Diffusion 3.5
205205

206206
```bash
207+
# Stable Diffusion 3
207208
python3 demo_txt2img_sd3.py "A vibrant street wall covered in colorful graffiti, the centerpiece spells \"SD3 MEDIUM\", in a storm of colors" --version sd3 --hf-token=$HF_TOKEN
209+
210+
# Stable Diffusion 3.5-medium
211+
python3 demo_txt2img_sd35.py "a beautiful photograph of Mt. Fuji during cherry blossom" --version=3.5-medium --denoising-steps=30 --guidance-scale 3.5 --hf-token=$HF_TOKEN
212+
213+
# Stable Diffusion 3.5-large
214+
python3 demo_txt2img_sd35.py "a beautiful photograph of Mt. Fuji during cherry blossom" --version=3.5-large --denoising-steps=30 --guidance-scale 3.5 --hf-token=$HF_TOKEN
208215
```
209216

210217
You can also specify an input image conditioning as shown below
211218

212219
```bash
213220
wget https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png -O dog-on-bench.png
214221

222+
# Stable Diffusion 3
215223
python3 demo_txt2img_sd3.py "dog wearing a sweater and a blue collar" --version sd3 --input-image dog-on-bench.png --hf-token=$HF_TOKEN
216224
```
217225

@@ -352,7 +360,7 @@ You can use the `--calibraton-dataset` flag to specify the path, which is set to
352360
python3 demo_img2img_flux.py "A robot made of exotic candies and chocolates of different kinds. The background is filled with confetti and celebratory gifts." --version="flux.1-dev-depth" --hf-token=$HF_TOKEN --guidance-scale 10 --control-image robot.png --bf16 --denoising-steps 30 --download-onnx-models
353361

354362
# FP8 using pre-exported ONNX models
355-
python3 demo_img2img_flux.py "A robot made of exotic candies" --version="flux.1-dev-depth" --hf-token=$HF_TOKEN --guidance-scale 10 --control-image robot.png --fp8 --denoising-steps 30 --download-onnx-models --build-static-batch
363+
python3 demo_img2img_flux.py "A robot made of exotic candies" --version="flux.1-dev-depth" --hf-token=$HF_TOKEN --guidance-scale 10 --control-image robot.png --fp8 --denoising-steps 30 --download-onnx-models --build-static-batch --quantization-level 4
356364

357365
# FP8 using native ONNX export
358366
rm -rf onnx/* engine/* && python3 demo_img2img_flux.py "A robot made of exotic candies" --version="flux.1-dev-depth" --hf-token=$HF_TOKEN --guidance-scale 10 --control-image robot.png --quantization-level 4 --fp8 --denoising-steps 30
@@ -368,13 +376,13 @@ python3 demo_img2img_flux.py "A robot made of exotic candies" --version="flux.1-
368376
python3 demo_img2img_flux.py "a robot made out of gold" --version="flux.1-dev-canny" --hf-token=$HF_TOKEN --guidance-scale 30 --control-image robot.png --bf16 --denoising-steps 30 --download-onnx-models
369377

370378
# FP8 using pre-exported ONNX models
371-
python3 demo_img2img_flux.py "a robot made out of gold" --version="flux.1-dev-canny" --hf-token=$HF_TOKEN --guidance-scale 30 --control-image robot.png --fp8 --denoising-steps 30 --download-onnx-models --build-static-batch
379+
python3 demo_img2img_flux.py "a robot made out of gold" --version="flux.1-dev-canny" --hf-token=$HF_TOKEN --guidance-scale 30 --control-image robot.png --fp8 --denoising-steps 30 --download-onnx-models --build-static-batch --quantization-level 4
372380

373381
# FP8 using native ONNX export
374382
rm -rf onnx/* engine/* && python3 demo_img2img_flux.py "a robot made out of gold" --version="flux.1-dev-canny" --hf-token=$HF_TOKEN --guidance-scale 30 --control-image robot.png --quantization-level 4 --fp8 --denoising-steps 30 --calibration-dataset {custom/dataset/path}
375383

376384
# FP4
377-
python3 demo_img2img_flux.py "a robot made out of gold" --version="flux.1-dev-canny" --hf-token=$HF_TOKEN --guidance-scale 30 --control-image robot.png --fp4 --denoising-steps 30 --download-onnx-models
385+
python3 demo_img2img_flux.py "a robot made out of gold" --version="flux.1-dev-canny" --hf-token=$HF_TOKEN --guidance-scale 30 --control-image robot.png --fp4 --denoising-steps 30 --download-onnx-models --build-static-batch
378386
```
379387

380388
#### 4. Generate an Image Using Flux LoRA

‎demo/Diffusion/demo_diffusion/dd_argparse.py

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -71,6 +71,8 @@ def add_arguments(parser):
7171
"xl-turbo",
7272
"svd-xt-1.1",
7373
"sd3",
74+
"3.5-medium",
75+
"3.5-large",
7476
"cascade",
7577
"flux.1-dev",
7678
"flux.1-schnell",
@@ -274,6 +276,7 @@ def process_pipeline_args(args: argparse.Namespace) -> Tuple[Dict[str, Any], Dic
274276
sm_version = device_info.major * 10 + device_info.minor
275277

276278
is_flux = args.version.startswith("flux")
279+
is_sd35 = args.version.startswith("3.5")
277280

278281
if args.height % 8 != 0 or args.width % 8 != 0:
279282
raise ValueError(
@@ -336,7 +339,6 @@ def override_quant_level(level: float, dtype_str: str):
336339
elif args.int8:
337340
override_quant_level(3.0, "INT8")
338341

339-
340342
if args.quantization_level == 3.0 and args.download_onnx_models:
341343
raise ValueError(
342344
"Transformer ONNX model for Quantization level 3 is not available for download. Please export the quantized Transformer model natively with the removal of --download-onnx-models."
@@ -366,7 +368,7 @@ def override_quant_level(level: float, dtype_str: str):
366368

367369
# Torch-fallback and Torch-inference
368370
if args.torch_fallback and not args.torch_inference:
369-
assert is_flux, "PyTorch Fallback is only supported for Flux pipelines"
371+
assert is_flux or is_sd35, "PyTorch Fallback is only supported for Flux and Stable Diffusion 3.5 pipelines."
370372
args.torch_fallback = args.torch_fallback.split(",")
371373

372374
if args.torch_fallback and args.torch_inference:
@@ -377,7 +379,7 @@ def override_quant_level(level: float, dtype_str: str):
377379

378380
# low-vram
379381
if args.low_vram:
380-
assert is_flux, "low-vram mode is only supported for Flux pipelines"
382+
assert is_flux or is_sd35, "low-vram mode is only supported for Flux and Stable Diffusion 3.5 pipelines."
381383

382384
# Pack arguments
383385
kwargs_init_pipeline = {

‎demo/Diffusion/demo_diffusion/engine.py

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -22,16 +22,17 @@
2222
from collections import OrderedDict, defaultdict
2323

2424
import numpy as np
25-
import onnx
2625
import tensorrt as trt
2726
import torch
2827
from cuda import cudart
29-
from onnx import numpy_helper
3028
from polygraphy.backend.common import bytes_from_path
3129
from polygraphy.backend.trt import (
3230
engine_from_bytes,
3331
)
3432

33+
import onnx
34+
from onnx import numpy_helper
35+
3536
TRT_LOGGER = trt.Logger(trt.Logger.ERROR)
3637

3738

‎demo/Diffusion/demo_diffusion/model/__init__.py

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -29,6 +29,7 @@
2929
from demo_diffusion.model.diffusion_transformer import (
3030
FluxTransformerModel,
3131
SD3_MMDiTModel,
32+
SD3TransformerModel,
3233
)
3334
from demo_diffusion.model.gan import VQGANModel
3435
from demo_diffusion.model.load import unload_torch_model
@@ -67,6 +68,7 @@
6768
# diffusion_transformer
6869
"SD3_MMDiTModel",
6970
"FluxTransformerModel",
71+
"SD3TransformerModel",
7072
# gan
7173
"VQGANModel",
7274
# lora

0 commit comments

Comments
 (0)