• Starting with the 25.01 release, Triton Inference Server supports Blackwell GPU architectures.

• Starting from 25.01, the vLLM container shipped by Triton is NVIDIA optimized. Users who wish to use the public version of vLLM can continue to build a Triton-vLLM container on their end.

• Fixed a bug when passing the correlation ID of string type to python_backend. Added datatype checks to correlation ID values.

• vLLM backend can now take advantage of the vLLM v0.6 performance improvement by communicating with the vLLM engine via ZMQ.

• GenAI-Perf now provides the exact input sequence length requested for synthetic text generation

• GenAI-Perf supports the creation of a prefix pool to emulate system prompts via --num-system-prompts and --system-prompt-length

• GenAI-Perf improved error visibility via returning more detailed errors when OpenAI frontends return an error or metric generation fails

• GenAI-Perf reports time-to-second-token and request count in its metrics

• GenAI-Perf allows the use of a custom tokenizer in its “compare” subcommand for comparing multiple profiles

• GenAI-Perf natively supports --request-count for sending a specific number of requests and --header for sending a list of headers with every request

• Model Analyzer functionality has been migrated to GenAI-Perf via the “analyze” subcommand, enabling the tool to sweep and find the optimal model configuration

• A bytes appending bug was fixed in GenAI-Perf, resulting in more accurate output sequence lengths for Triton

• A segmentation fault related to DCGM and NSCQ may be encountered during server shutdown on NVSwitch systems. A possible workaround for this issue is to disable the collection of GPU metrics tritonserver --allow-gpu-metrics false ...

• vLLM backend currently does not take advantage of the vLLM v0.6 performance improvement when metrics are enabled.

• Please note, that the vllm version provided in 25.01 container is 0.6.3.post1. Due to some issues with vllm library versioning, vllm.__version__ displays 0.6.3.

• Incorrect results are known to occur when using TensorRT (TRT) Backend for inference using int8 data type for I/O on the Blackwell GPU architecture.

• When running Torch TRT models, the output may differ from running the same model on a previous release.

• When using TensorRT models, if auto-complete configuration is disabled and is_non_linear_format_io:true for reformat-free tensors is not provided in the model configuration, the model may not load successfully.

• 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.

• Triton Inference Server with vLLM backend currently does not support running vLLM models with tensor parallelism sizes greater than 1 and the default "distributed_executor_backend" setting when using explicit model control mode. In attempt to load a vllm model (tp > 1) in explicit mode, users could potentially see failure at initialize step: could not acquire lock for <_io.BufferedWriter name='<stdout>'> at interpreter shutdown, possibly due to daemon threads. For the default model control mode, after server shutdown, vllm related sub-processes are not killed. Related vllm issue: vllm-project/vllm#6766 . Please specify "distributed_executor_backend":"ray" in the model.json when deploying vllm models with tensor parallelism > 1.

• When loading models with file override, multiple model configuration files are not supported. Users must provide the model configuration by setting parameter "config" : "<JSON>" instead of custom configuration file in the following format: "file:configs/<model-config-name>.pbtxt" : "<base64-encoded-file-content>".

• 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.

• 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. NVIDIA recommends 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: https://github.com/pytorch/pytorch/issues/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.

• 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

Ubuntu 24.04 builds of the client libraries and examples are included in this release in the attached v2.54.0_ubuntu2404.clients.tar.gz file. The SDK is also available for as an Ubuntu 24.04 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.

A beta release of Triton for Windows is provided in the attached file: tritonserver2.54.0-win.zip. This is a beta release so functionality is limited and performance is not optimized. Additional features and improved performance will be provided in future releases. Specifically in this release:

• HTTP/REST and GRPC endpoints are supported.

*ONNX models are supported by the ONNX Runtime backend. The ONNX Runtime version is 1.20.1. The CPU, CUDA, and TensorRT execution providers are supported. The OpenVINO execution provider is not supported.

• OpenVINO models are supported. The OpenVINO version is 2024.5.0.

• Prometheus metrics endpoint is not supported.

• System and CUDA shared memory are not supported.

Known Issues

• In our internal testing, we observed large latency in retrieving inference results from HTTP client. We recommend using gRPC to circumvent this issue.

To use the Windows version of Triton, you must install all the necessary dependencies on your Windows system. These dependencies are available in the Dockerfile.win10.min. The Dockerfile includes the following CUDA-related components:

• Python 3.12.3

• CUDA 12.6.3

• cuDNN 9.6.0.74

• TensorRT 10.7.0.23

A release of Triton for IGX is provided in the attached tar file: tritonserver2.54.0-igpu.tgz.

• This release supports TensorFlow 2.17.0, TensorRT 10.8.0.40, Onnx Runtime 1.20.1, PyTorch 2.6.0a0+ecf3bae40a, Python 3.12 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-2.54.0-py3-none-manylinux2014_aarch64.whl[all]

The Triton TensorRT-LLM container is built from the 25.01 image nvcr.io/nvidia/tritonserver:25.01-py3-min. Please refer to the support matrix for all dependency versions related to 25.01. However, the packages listed below have different versions than those specified in the support matrix.