Triton Inference Server

The Triton Inference Server provides a cloud inferencing solution optimized for both CPUs and GPUs. The server provides an inference service via an HTTP or GRPC endpoint, allowing remote clients to request inferencing for any model being managed by the server. For edge deployments, Triton Server is also available as a shared library with an API that allows the full functionality of the server to be included directly in an application.

New Features and Improvements

• The TensorRT Backend now supports the BF16 datatype.

• A new tutorial on auto-scaling and load balancing TensorRT-LLM model deployments with Triton Inference Server has been released: https://github.com/triton-inference-server/tutorials/tree/main/Deployment/Kubernetes/TensorRT-LLM_Autoscaling_and_Load_Balancing

• A compare subcommand has been added to GenAi-Perf to allow comparison across multiple runs

• Multi-LoRA and multi-model support in GenAI-Perf

• Custom visualizations in GenAI-Perf

• A fixed request count can now be requested from Perf Analyzer

• Ensemble top-level response caching support in Perf Analyzer

• Added --enable-peer-access to control trying to enable GPU peer access on triton startup.. Default is TRUE.

• Python models in default mode may send its response using the InferenceResponseSender similarly to models in decoupled mode.

• Addressed an issue where Triton would cease processing gRPC requests after receiving multiple cancellation requests.

Known Issues

• Starting in 24.06, if you use Triton's iGPU container you might encounter this error message when loading TensorRT models built with the 24.06 TensorRT iGPU container: Serialization (Serialization assertion stdVersionRead == serializationVersion failed.Version tag does not match. Note: Current Version: 236, Serialized Engine Version: 237). If this happens you can rebuild your iGPU models with the 24.04 TensorRT iGPU container and then run them in the Triton 24.06 iGPU container.

• Triton TensorRT-LLM Backend container image uses TensorRT-LLM version 0.9.0 and built out of nvcr.io/nvidia/pytorch:24.03-py3

• When using Python models in decoupled mode, users need to ensure that the ResponseSender goes out of scope or is properly cleaned up before unloading the model to guarantee that the unloading process executes correctly.

• Restart support was temporarily removed for Python models.

• TensorRT v10 does not support implicit batching. As a result, Triton no longer supports TensorRT models with implicit batch dimensions.

• Since TensorRT v10 no longer supports implicit batch, Tritonserver will not be able to load existing TF-TRT models that use implicit batch. Therefore, we need to build TF-TRT models with dynamic batch support.

• Multiple model configuration files are not supported by loading models with file override. Users still need to provide the model configuration by setting parameter "config" : "<JSON>" instead of custom configuration file "file:configs/<model-config-name>.pbtxt" : "<base64-encoded-file-content>"

• Perf Analyzer no longer supports --trace-file option.

• The TensorRT-LLM backend provides limited support of Triton extensions and features.

• The TensorRT-LLM backend may core dump on server shutdown. This impacts server teardown only and will not impact inferencing.

• The Java CAPI is known to have intermittent segfaults we’re looking for a root cause.

• Some systems which implement malloc() may not release memory back to the operating system right away causing a false memory leak. This can be mitigated by using a different malloc implementation. Tcmalloc and jemalloc are installed in the Triton container and can be used by specifying the library in LD_PRELOAD. We recommend experimenting with both tcmalloc and jemalloc to determine which one works better for your use case.

• Auto-complete may cause an increase in server start time. To avoid a start time increase, users can provide the full model configuration and launch the server with --disable-auto-complete-config.

• Auto-complete does not support PyTorch models due to lack of metadata in the model. It can only verify that the number of inputs and the input names matches what is specified in the model configuration. There is no model metadata about the number of outputs and datatypes. Related PyTorch bug: pytorch/pytorch#38273

• Triton Client PIP wheels for ARM SBSA are not available from PyPI and pip will install an incorrect Jetson version of Triton Client library for Arm SBSA. The correct client wheel file can be pulled directly from the Arm SBSA SDK image and manually installed.

• Traced models in PyTorch seem to create overflows when int8 tensor values are transformed to int32 on the GPU. Refer to pytorch/pytorch#66930 for more information.

• Triton cannot retrieve GPU metrics with MIG-enabled GPU devices (A100 and A30).

• Triton metrics might not work if the host machine is running a separate DCGM agent on bare-metal or in a container.

• When cloud storage (AWS, GCS, AZURE) is used as a model repository and a model has multiple versions, Triton creates an extra local copy of the cloud model’s folder in the temporary directory, which is deleted upon server’s shutdown.

• Python backend support for Windows is limited and does not currently support the following features:

  • GPU tensors
  • CPU and GPU-related metrics
  • Custom execution environments
  • The model load/unload APIs

Client Libraries and Examples

Ubuntu 22.04 builds of the client libraries and examples are included in this release in the attached v2.47.0_ubuntu22.04.clients.tar.gz file. The SDK is also available for as an Ubuntu 22.04 based NGC Container. The SDK container includes the client libraries and examples, Performance Analyzer and Model Analyzer. Some components are also available in the tritonclient pip package. See Getting the Client Libraries for more information on each of these options.

For Windows, the client libraries and some examples are available in the attached tritonserver2.47.0-sdk-win.zip file.

Windows Support

Jetson iGPU Support

A release of Triton for IGX is provided in the attached tar file: tritonserver2.47.0-igpu.tgz.

• This release supports TensorFlow 2.16.1, TensorRT 8.6.2.3, Onnx Runtime 1.18.0, PyTorch 2.4.0a0+f70bd71a48, Python 3.10 and as well as ensembles.
• ONNX Runtime backend does not support the OpenVINO and TensorRT execution providers. The CUDA execution provider is in Beta.
• System shared memory is supported on Jetson. CUDA shared memory is not supported.
• GPU metrics, GCS storage, S3 storage and Azure storage are not supported.

The tar file contains the Triton server executable and shared libraries and also the C++ and Python client libraries and examples. For more information on how to install and use Triton on JetPack refer to jetson.md.

The wheel for the Python client library is present in the tar file and can be installed by running the following command:

python3 -m pip install --upgrade clients/python/tritonclient-2.47.0-py3-none-manylinux2014_aarch64.whl[all]