TensorRT 10.13.3 OSS Release (#4570)

Signed-off-by: Kevin Chen <[email protected]>

Author: kevinch-nv

CHANGELOG.md

@@ -1,5 +1,11 @@
 # TensorRT OSS Release Changelog
 
+## 10.13.3 GA - 2025-9-8
+- Added support for TensorRT API Capture and Replay feature, see the [developer guide](https://docs.nvidia.com/deeplearning/tensorrt/latest/inference-library/advanced.html) for more information.
+- Demo changes
+  - Added support for Flux Kontext pipeline.
+
+
 ## 10.13.2 GA - 2025-8-18
 - Added support for CUDA 13.0, dropped support for CUDA 11.X
 - Dropped support for Ubuntu 20.04

CMakeLists.txt

@@ -66,22 +66,37 @@ endif()
 
 set(CMAKE_SKIP_BUILD_RPATH True)
 
-# Set CUDA architectures before enabling CUDA language to avoid detection issues in containers
-if (DEFINED GPU_ARCHS AND NOT GPU_ARCHS STREQUAL "")
-  message(STATUS "GPU_ARCHS defined as ${GPU_ARCHS}. Setting CUDA architectures for SM ${GPU_ARCHS}")
+# CUDA targets
+
+if (DEFINED GPU_ARCHS)
+  message(STATUS "GPU_ARCHS defined as ${GPU_ARCHS}. Generating CUDA code for SM ${GPU_ARCHS}")
   separate_arguments(GPU_ARCHS)
   foreach(SM IN LISTS GPU_ARCHS)
     list(APPEND CMAKE_CUDA_ARCHITECTURES "${SM}")
   endforeach()
 else()
-  # Set default architectures for container builds where auto-detection fails
-  set(CMAKE_CUDA_ARCHITECTURES 75 80 86 87 89 90)
+  list(APPEND CMAKE_CUDA_ARCHITECTURES 75 80 86 87 89 90)
 
   if(CUDA_VERSION VERSION_GREATER_EQUAL 12.8)
       list(APPEND CMAKE_CUDA_ARCHITECTURES 100 120)
   endif()
 
-  message(STATUS "Setting default CUDA architectures for container build: ${CMAKE_CUDA_ARCHITECTURES}")
+  message(STATUS "GPU_ARCHS is not defined. Generating CUDA code for default SMs: ${CMAKE_CUDA_ARCHITECTURES}")
+endif()
+set(BERT_GENCODES)
+# Generate SASS for each architecture
+foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
+    if (${arch} GREATER_EQUAL 75)
+        set(BERT_GENCODES "${BERT_GENCODES} -gencode arch=compute_${arch},code=sm_${arch}")
+    endif()
+    set(GENCODES "${GENCODES} -gencode arch=compute_${arch},code=sm_${arch}")
+endforeach()
+
+# Generate PTX for the last architecture in the list.
+list(GET CMAKE_CUDA_ARCHITECTURES -1 LATEST_SM)
+set(GENCODES "${GENCODES} -gencode arch=compute_${LATEST_SM},code=compute_${LATEST_SM}")
+if (${LATEST_SM} GREATER_EQUAL 75)
+    set(BERT_GENCODES "${BERT_GENCODES} -gencode arch=compute_${LATEST_SM},code=compute_${LATEST_SM}")
 endif()
 
 project(TensorRT
@@ -195,21 +210,6 @@ endif()
 set(CUDA_LIBRARIES ${CUDART_LIB})
 
 ############################################################################################
-set(BERT_GENCODES)
-# Generate SASS for each architecture
-foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
-    if (${arch} GREATER_EQUAL 75)
-        set(BERT_GENCODES "${BERT_GENCODES} -gencode arch=compute_${arch},code=sm_${arch}")
-    endif()
-    set(GENCODES "${GENCODES} -gencode arch=compute_${arch},code=sm_${arch}")
-endforeach()
-
-# Generate PTX for the last architecture in the list.
-list(GET CMAKE_CUDA_ARCHITECTURES -1 LATEST_SM)
-set(GENCODES "${GENCODES} -gencode arch=compute_${LATEST_SM},code=compute_${LATEST_SM}")
-if (${LATEST_SM} GREATER_EQUAL 75)
-    set(BERT_GENCODES "${BERT_GENCODES} -gencode arch=compute_${LATEST_SM},code=compute_${LATEST_SM}")
-endif()
 
 if(NOT MSVC)
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr -Xcompiler -Wno-deprecated-declarations")

README.md

@@ -32,7 +32,7 @@ To build the TensorRT-OSS components, you will first need the following software
 
 **TensorRT GA build**
 
-- TensorRT v10.13.2.6
+- TensorRT v10.13.3.9
   - Available from direct download links listed below
 
 **System Packages**
@@ -86,24 +86,24 @@ To build the TensorRT-OSS components, you will first need the following software
 
    Else download and extract the TensorRT GA build from [NVIDIA Developer Zone](https://developer.nvidia.com) with the direct links below:
 
-   - [TensorRT 10.13.2.6 for CUDA 13.0, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.x86_64-gnu.cuda-13.0.tar.gz)
-   - [TensorRT 10.13.2.6 for CUDA 12.9, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.x86_64-gnu.cuda-12.9.tar.gz)
-   - [TensorRT 10.13.2.6 for CUDA 13.0, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/zip/TensorRT-10.13.2.6.Windows.win10.cuda-13.0.zip)
-   - [TensorRT 10.13.2.6 for CUDA 12.9, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/zip/TensorRT-10.13.2.6.Windows.win10.cuda-12.9.zip)
+   - [TensorRT 10.13.3.9 for CUDA 13.0, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.x86_64-gnu.cuda-13.0.tar.gz)
+   - [TensorRT 10.13.3.9 for CUDA 12.9, Linux x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/tars/TensorRT-10.13.3.9.Linux.x86_64-gnu.cuda-12.9.tar.gz)
+   - [TensorRT 10.13.3.9 for CUDA 13.0, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/zip/TensorRT-10.13.3.9.Windows.win10.cuda-13.0.zip)
+   - [TensorRT 10.13.3.9 for CUDA 12.9, Windows x86_64](https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.3/zip/TensorRT-10.13.3.9.Windows.win10.cuda-12.9.zip)
 
    **Example: Ubuntu 22.04 on x86-64 with cuda-13.0**
 
    ```bash
    cd ~/Downloads
-   tar -xvzf TensorRT-10.13.2.6.Linux.x86_64-gnu.cuda-13.0.tar.gz
-   export TRT_LIBPATH=`pwd`/TensorRT-10.13.2.6
+   tar -xvzf TensorRT-10.13.3.9.Linux.x86_64-gnu.cuda-13.0.tar.gz
+   export TRT_LIBPATH=`pwd`/TensorRT-10.13.3.9
    ```
 
    **Example: Windows on x86-64 with cuda-12.9**
 
    ```powershell
-   Expand-Archive -Path TensorRT-10.13.2.6.Windows.win10.cuda-12.9.zip
-   $env:TRT_LIBPATH="$pwd\TensorRT-10.13.2.6\lib"
+   Expand-Archive -Path TensorRT-10.13.3.9.Windows.win10.cuda-12.9.zip
+   $env:TRT_LIBPATH="$pwd\TensorRT-10.13.3.9\lib"
    ```
 
 ## Setting Up The Build Environment
@@ -124,16 +124,16 @@ For Linux platforms, we recommend that you generate a docker container for build
    ./docker/build.sh --file docker/rockylinux8.Dockerfile --tag tensorrt-rockylinux8-cuda12.9
    ```
 
-   **Example: Ubuntu 22.04 cross-compile for Jetson (aarch64) with cuda-12.9 (JetPack SDK)**
+   **Example: Ubuntu 24.04 cross-compile for Jetson (aarch64) with cuda-13.0 (JetPack SDK)**
 
    ```bash
-   ./docker/build.sh --file docker/ubuntu-cross-aarch64.Dockerfile --tag tensorrt-jetpack-cuda12.9
+   ./docker/build.sh --file docker/ubuntu-cross-aarch64.Dockerfile --tag tensorrt-jetpack-cuda13.0
    ```
 
-   **Example: Ubuntu 22.04 on aarch64 with cuda-12.9**
+   **Example: Ubuntu 24.04 on aarch64 with cuda-13.0**
 
    ```bash
-   ./docker/build.sh --file docker/ubuntu-22.04-aarch64.Dockerfile --tag tensorrt-aarch64-ubuntu22.04-cuda12.9
+   ./docker/build.sh --file docker/ubuntu-24.04-aarch64.Dockerfile --tag tensorrt-aarch64-ubuntu24.04-cuda13.0
    ```
 
 2. #### Launch the TensorRT-OSS build container.
@@ -146,12 +146,13 @@ For Linux platforms, we recommend that you generate a docker container for build
    > <br> 2. [NVIDIA Container Toolkit](#prerequisites) is required for GPU access (running TensorRT applications) inside the build container.
    > <br> 3. `sudo` password for Ubuntu build containers is 'nvidia'.
    > <br> 4. Specify port number using `--jupyter <port>` for launching Jupyter notebooks.
+   > <br> 5. Write permission to this folder is required as this folder will be mounted inside the docker container for uid:gid of 1000:1000.
 
 ## Building TensorRT-OSS
 
 - Generate Makefiles and build
 
-  **Example: Linux (x86-64) build with default cuda-12.9**
+  **Example: Linux (x86-64) build with default cuda-13.0**
 
   ```bash
   cd $TRT_OSSPATH
@@ -160,7 +161,7 @@ For Linux platforms, we recommend that you generate a docker container for build
   make -j$(nproc)
   ```
 
-  **Example: Linux (aarch64) build with default cuda-12.9**
+  **Example: Linux (aarch64) build with default cuda-13.0**
 
   ```bash
   cd $TRT_OSSPATH
@@ -169,27 +170,27 @@ For Linux platforms, we recommend that you generate a docker container for build
   make -j$(nproc)
   ```
 
-  **Example: Native build on Jetson (aarch64) with cuda-12.9**
+  **Example: Native build on Jetson Thor (aarch64) with cuda-13.0**
 
   ```bash
   cd $TRT_OSSPATH
   mkdir -p build && cd build
-  cmake .. -DTRT_LIB_DIR=$TRT_LIBPATH -DTRT_OUT_DIR=`pwd`/out -DTRT_PLATFORM_ID=aarch64 -DCUDA_VERSION=12.9
+  cmake .. -DTRT_LIB_DIR=$TRT_LIBPATH -DTRT_OUT_DIR=`pwd`/out -DTRT_PLATFORM_ID=aarch64 -DGPU_ARCHS=110
   CC=/usr/bin/gcc make -j$(nproc)
   ```
 
   > NOTE: C compiler must be explicitly specified via CC= for native aarch64 builds of protobuf.
 
-  **Example: Ubuntu 22.04 Cross-Compile for Jetson (aarch64) with cuda-12.9 (JetPack)**
+  **Example: Ubuntu 24.04 Cross-Compile for Jetson Thor (aarch64) with cuda-13.0 (JetPack)**
 
   ```bash
   cd $TRT_OSSPATH
   mkdir -p build && cd build
-  cmake .. -DCMAKE_TOOLCHAIN_FILE=$TRT_OSSPATH/cmake/toolchains/cmake_aarch64.toolchain -DCUDA_VERSION=12.9 -DCUDNN_LIB=/pdk_files/cudnn/usr/lib/aarch64-linux-gnu/libcudnn.so -DCUBLAS_LIB=/usr/local/cuda-12.9/targets/aarch64-linux/lib/stubs/libcublas.so -DCUBLASLT_LIB=/usr/local/cuda-12.9/targets/aarch64-linux/lib/stubs/libcublasLt.so -DTRT_LIB_DIR=/pdk_files/tensorrt/lib
+  cmake .. -DTRT_LIB_DIR=$TRT_LIBPATH -DCMAKE_TOOLCHAIN_FILE=$TRT_OSSPATH/cmake/toolchains/cmake_aarch64_cross.toolchain -DGPU_ARCHS=110
   make -j$(nproc)
   ```
 
-  **Example: Native builds on Windows (x86) with cuda-12.9**
+  **Example: Native builds on Windows (x86) with cuda-13.0**
 
   ```bash
   cd $TRT_OSSPATH

VERSION

@@ -1 +1 @@
-10.13.2.6
+10.13.3.9

demo/Diffusion/README.md

@@ -7,7 +7,7 @@ This demo application ("demoDiffusion") showcases the acceleration of Stable Dif
 ### Clone the TensorRT OSS repository
 
 ```bash
-git clone [email protected]:NVIDIA/TensorRT.git -b release/10.13.2 --single-branch
+git clone [email protected]:NVIDIA/TensorRT.git -b release/10.13.3 --single-branch
 cd TensorRT
 ```
 
@@ -16,45 +16,37 @@ cd TensorRT
 Install nvidia-docker using [these intructions](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker).
 
 ```bash
-docker run --rm -it --gpus all -v $PWD:/workspace nvcr.io/nvidia/pytorch:25.01-py3 /bin/bash
+docker run --rm -it --gpus all -v $PWD:/workspace nvcr.io/nvidia/pytorch:25.08-py3 /bin/bash
 ```
 
-NOTE: The demo supports CUDA>=12
+NOTE: The demo supports CUDA>=12.0
 
-### Install latest TensorRT release
+### Install the required packages
 
 ```bash
-python3 -m pip install --upgrade pip
-pip install --pre tensorrt-cu12
+source setup.sh
 ```
 
 Check your installed version using:
 `python3 -c 'import tensorrt;print(tensorrt.__version__)'`
 
 > NOTE: Alternatively, you can download and install TensorRT packages from [NVIDIA TensorRT Developer Zone](https://developer.nvidia.com/tensorrt).
 
-### Install required packages
-
-```bash
-export TRT_OSSPATH=/workspace
-cd $TRT_OSSPATH/demo/Diffusion
-pip3 install -r requirements.txt
-```
 
 > NOTE: demoDiffusion has been tested on systems with NVIDIA H100, A100, L40, T4, and RTX4090 GPUs, and the following software configuration.
 
 ```
-diffusers           0.31.0
-onnx                1.15.0
+diffusers           0.35.0
+onnx                1.18.0
 onnx-graphsurgeon   0.5.2
-onnxruntime         1.16.3
-polygraphy          0.49.9
-tensorrt            10.13.2.6
+onnxruntime         1.19.2
+polygraphy          0.49.22
+tensorrt            10.13.3.9
 tokenizers          0.13.3
-torch               2.2.0
-transformers        4.42.2
+torch               2.8.0a0+5228986c39.nv25.6
+transformers        4.52.4
 controlnet-aux      0.0.6
-nvidia-modelopt     0.15.1
+nvidia-modelopt     0.31.0
 ```
 
 # Running demoDiffusion
@@ -210,7 +202,7 @@ Run the command below to generate an image using Stable Diffusion 3 and Stable D
 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
 
 # Stable Diffusion 3.5-medium
-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=$HF_TOKEN --bf16
+python3 demo_txt2img_sd35.py "a beautiful photograph of Mt. Fuji during cherry blossom" --version=3.5-medium --denoising-steps=30 --guidance-scale 3.5 --hf-token=$HF_TOKEN --bf16 --download-onnx-models
 
 # Stable Diffusion 3.5-large
 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 --bf16 --download-onnx-models
@@ -234,13 +226,13 @@ Note that a denosing-percentage is applied to the number of denoising-steps when
 
 ```bash
 # Depth
-python3 demo_controlnet_sd35.py "a photo of a man" --controlnet-type depth --hf-token=$HF_TOKEN --denoising-steps 40 --guidance-scale 4.5 --bf16
+python3 demo_controlnet_sd35.py "a photo of a man" --controlnet-type depth --hf-token=$HF_TOKEN --denoising-steps 40 --guidance-scale 4.5 --bf16 --download-onnx-models
 
 # Canny
-python3 demo_controlnet_sd35.py "A Night time photo taken by Leica M11, portrait of a Japanese woman in a kimono, looking at the camera, Cherry blossoms" --controlnet-type canny --hf-token=$HF_TOKEN --denoising-steps 60 --guidance-scale 3.5 --bf16
+python3 demo_controlnet_sd35.py "A Night time photo taken by Leica M11, portrait of a Japanese woman in a kimono, looking at the camera, Cherry blossoms" --controlnet-type canny --hf-token=$HF_TOKEN --denoising-steps 60 --guidance-scale 3.5 --bf16 --download-onnx-models
 
 # Blur
-python3 demo_controlnet_sd35.py "generated ai art, a tiny, lost rubber ducky in an action shot close-up, surfing the humongous waves, inside the tube, in the style of Kelly Slater" --controlnet-type blur --hf-token=$HF_TOKEN --denoising-steps 60 --guidance-scale 3.5 --bf16
+python3 demo_controlnet_sd35.py "generated ai art, a tiny, lost rubber ducky in an action shot close-up, surfing the humongous waves, inside the tube, in the style of Kelly Slater" --controlnet-type blur --hf-token=$HF_TOKEN --denoising-steps 60 --guidance-scale 3.5 --bf16 --download-onnx-models
 ```
 
 ### Generate a video guided by an initial image using Stable Video Diffusion
@@ -414,6 +406,21 @@ python3 demo_txt2img_flux.py "A painting of a barista creating an intricate latt
 # FP8
 python3 demo_txt2img_flux.py "A painting of a barista creating an intricate latte art design, with the 'Coffee Creations' logo skillfully formed within the latte foam. In a watercolor style, AQUACOLTOK. White background." --hf-token=$HF_TOKEN --lora-path "SebastianBodza/flux_lora_aquarel_watercolor" --lora-weight 1.0 --onnx-dir=onnx-flux-lora --engine-dir=engine-flux-lora --fp8
 ```
+
+#### 5. Edit an Image using Flux Kontext
+
+```bash
+wget https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png
+
+# BF16
+python3 demo_img2img_flux.py "Add a hat to the cat" --version="flux.1-kontext-dev" --hf-token=$HF_TOKEN --guidance-scale 2.5 --kontext-image cat.png --denoising-steps 28 --bf16 --onnx-dir onnx-kontext --engine-dir engine-kontext --download-onnx-models
+
+# FP8
+python3 demo_img2img_flux.py "Add a hat to the cat" --version="flux.1-kontext-dev" --hf-token=$HF_TOKEN --guidance-scale 2.5 --kontext-image cat.png --denoising-steps 28 --fp8 --onnx-dir onnx-kontext-fp8 --engine-dir engine-kontext-fp8 --download-onnx-models --quantization-level 4
+
+# FP4
+python3 demo_img2img_flux.py "Add a hat to the cat" --version="flux.1-kontext-dev" --hf-token=$HF_TOKEN --guidance-scale 2.5 --kontext-image cat.png --denoising-steps 28 --fp4 --onnx-dir onnx-kontext-fp4 --engine-dir engine-kontext-fp4 --download-onnx-models
+```
 ---
 
 #### 5. Export ONNX Models Only (Skip Inference)

demo/Diffusion/demo_controlnet.py

@@ -78,49 +78,49 @@ def parseArgs():
                     raise ValueError(
                         f"This demo supports ControlNets for v1.5 and SDXL base pipelines only. Version provided: {args.version}"
                     )
-                input_images.append(canny_image.resize((args.height, args.width)))
+                input_images.append(canny_image.resize((args.width, args.height)))
             elif controlnet == "normal":
                 normal_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-normal/resolve/main/images/toy.png"
                 )
                 normal_image = controlnet_aux.NormalBaeDetector.from_pretrained("lllyasviel/Annotators")(normal_image)
-                input_images.append(normal_image.resize((args.height, args.width)))
+                input_images.append(normal_image.resize((args.width, args.height)))
             elif controlnet == "depth":
                 depth_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-depth/resolve/main/images/stormtrooper.png"
                 )
                 depth_image = controlnet_aux.LeresDetector.from_pretrained("lllyasviel/Annotators")(depth_image)
-                input_images.append(depth_image.resize((args.height, args.width)))
+                input_images.append(depth_image.resize((args.width, args.height)))
             elif controlnet == "hed":
                 hed_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-hed/resolve/main/images/man.png"
                 )
                 hed_image = controlnet_aux.HEDdetector.from_pretrained("lllyasviel/Annotators")(hed_image)
-                input_images.append(hed_image.resize((args.height, args.width)))
+                input_images.append(hed_image.resize((args.width, args.height)))
             elif controlnet == "mlsd":
                 mlsd_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-mlsd/resolve/main/images/room.png"
                 )
                 mlsd_image = controlnet_aux.MLSDdetector.from_pretrained("lllyasviel/Annotators")(mlsd_image)
-                input_images.append(mlsd_image.resize((args.height, args.width)))
+                input_images.append(mlsd_image.resize((args.width, args.height)))
             elif controlnet == "openpose":
                 openpose_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-openpose/resolve/main/images/pose.png"
                 )
                 openpose_image = controlnet_aux.OpenposeDetector.from_pretrained("lllyasviel/Annotators")(openpose_image)
-                input_images.append(openpose_image.resize((args.height, args.width)))
+                input_images.append(openpose_image.resize((args.width, args.height)))
             elif controlnet == "scribble":
                 scribble_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-scribble/resolve/main/images/bag.png"
                 )
                 scribble_image = controlnet_aux.HEDdetector.from_pretrained("lllyasviel/Annotators")(scribble_image, scribble=True)
-                input_images.append(scribble_image.resize((args.height, args.width)))
+                input_images.append(scribble_image.resize((args.width, args.height)))
             elif controlnet == "seg":
                 seg_image = image_module.download_image(
                     "https://huggingface.co/lllyasviel/sd-controlnet-seg/resolve/main/images/house.png"
                 )
                 seg_image = controlnet_aux.SamDetector.from_pretrained("ybelkada/segment-anything", subfolder="checkpoints")(seg_image)
-                input_images.append(seg_image.resize((args.height, args.width)))
+                input_images.append(seg_image.resize((args.width, args.height)))
             else:
                 raise ValueError(f"You should implement the conditonal image of this controlnet: {controlnet}")
     assert len(input_images) > 0