Skip to content

Commit 2d77fd0

Browse files
mc-nvmc-nv
authored
Update README for 22.12 release (#5172)
* Update README for 22.12 release
1 parent 3cfeca2 commit 2d77fd0

File tree

2 files changed

+304
-2
lines changed

2 files changed

+304
-2
lines changed

‎README.md

+213-2
Original file line numberDiff line numberDiff line change
@@ -28,5 +28,216 @@
2828

2929
# Triton Inference Server
3030

31-
**NOTE: You are currently on the r22.12 branch which tracks stabilization
32-
towards the next release. This branch is not usable during stabilization.**
31+
[![License](https://img.shields.io/badge/License-BSD3-lightgrey.svg)](https://opensource.org/licenses/BSD-3-Clause)
32+
----
33+
Triton Inference Server is an open source inference serving software that
34+
streamlines AI inferencing. Triton enables teams to deploy any AI model from
35+
multiple deep learning and machine learning frameworks, including TensorRT,
36+
TensorFlow, PyTorch, ONNX, OpenVINO, Python, RAPIDS FIL, and more. Triton
37+
supports inference across cloud, data center,edge and embedded devices on NVIDIA
38+
GPUs, x86 and ARM CPU, or AWS Inferentia. Triton delivers optimized performance
39+
for many query types, including real time, batched, ensembles and audio/video
40+
streaming.
41+
42+
Major features include:
43+
44+
- [Supports multiple deep learning
45+
frameworks](https://github.com/triton-inference-server/backend#where-can-i-find-all-the-backends-that-are-available-for-triton)
46+
- [Supports multiple machine learning
47+
frameworks](https://github.com/triton-inference-server/fil_backend)
48+
- [Concurrent model
49+
execution](docs/user_guide/architecture.md#concurrent-model-execution)
50+
- [Dynamic batching](docs/user_guide/model_configuration.md#dynamic-batcher)
51+
- [Sequence batching](docs/user_guide/model_configuration.md#sequence-batcher) and
52+
[implicit state management](docs/user_guide/architecture.md#implicit-state-management)
53+
for stateful models
54+
- Provides [Backend API](https://github.com/triton-inference-server/backend) that
55+
allows adding custom backends and pre/post processing operations
56+
- Model pipelines using
57+
[Ensembling](docs/user_guide/architecture.md#ensemble-models) or [Business
58+
Logic Scripting
59+
(BLS)](https://github.com/triton-inference-server/python_backend#business-logic-scripting)
60+
- [HTTP/REST and GRPC inference
61+
protocols](docs/customization_guide/inference_protocols.md) based on the community
62+
developed [KServe
63+
protocol](https://github.com/kserve/kserve/tree/master/docs/predict-api/v2)
64+
- A [C API](docs/customization_guide/inference_protocols.md#in-process-triton-server-api) and
65+
[Java API](docs/customization_guide/inference_protocols.md#java-bindings-for-in-process-triton-server-api)
66+
allow Triton to link directly into your application for edge and other in-process use cases
67+
- [Metrics](docs/user_guide/metrics.md) indicating GPU utilization, server
68+
throughput, server latency, and more
69+
70+
Join the [Triton and TensorRT community](https://www.nvidia.com/en-us/deep-learning-ai/triton-tensorrt-newsletter/) and
71+
stay current on the latest product updates, bug fixes, content, best practices,
72+
and more. Need enterprise support? NVIDIA global support is available for Triton
73+
Inference Server with the
74+
[NVIDIA AI Enterprise software suite](https://www.nvidia.com/en-us/data-center/products/ai-enterprise/).
75+
76+
## Serve a Model in 3 Easy Steps
77+
78+
```bash
79+
# Step 1: Create the example model repository
80+
git clone -b r22.12 https://github.com/triton-inference-server/server.git
81+
cd server/docs/examples
82+
./fetch_models.sh
83+
84+
# Step 2: Launch triton from the NGC Triton container
85+
docker run --gpus=1 --rm --net=host -v ${PWD}/model_repository:/models nvcr.io/nvidia/tritonserver:22.12-py3 tritonserver --model-repository=/models
86+
87+
# Step 3: Sending an Inference Request
88+
# In a separate console, launch the image_client example from the NGC Triton SDK container
89+
docker run -it --rm --net=host nvcr.io/nvidia/tritonserver:22.12-py3-sdk
90+
/workspace/install/bin/image_client -m densenet_onnx -c 3 -s INCEPTION /workspace/images/mug.jpg
91+
92+
# Inference should return the following
93+
Image '/workspace/images/mug.jpg':
94+
15.346230 (504) = COFFEE MUG
95+
13.224326 (968) = CUP
96+
10.422965 (505) = COFFEEPOT
97+
```
98+
Please read the [QuickStart](docs/getting_started/quickstart.md) guide for additional information
99+
regarding this example. The quickstart guide also contains an example of how to launch Triton on [CPU-only systems](docs/getting_started/quickstart.md#run-on-cpu-only-system). New to Triton and wondering where to get started? Watch the [Getting Started video](https://youtu.be/NQDtfSi5QF4).
100+
101+
## Examples and Tutorials
102+
103+
Check out [NVIDIA LaunchPad](https://www.nvidia.com/en-us/data-center/products/ai-enterprise-suite/trial/)
104+
for free access to a set of hands-on labs with Triton Inference Server hosted on
105+
NVIDIA infrastructure.
106+
107+
Specific end-to-end examples for popular models, such as ResNet, BERT, and DLRM
108+
are located in the
109+
[NVIDIA Deep Learning Examples](https://github.com/NVIDIA/DeepLearningExamples)
110+
page on GitHub. The
111+
[NVIDIA Developer Zone](https://developer.nvidia.com/nvidia-triton-inference-server)
112+
contains additional documentation, presentations, and examples.
113+
114+
## Documentation
115+
116+
### Build and Deploy
117+
118+
The recommended way to build and use Triton Inference Server is with Docker
119+
images.
120+
121+
- [Install Triton Inference Server with Docker containers](docs/customization_guide/build.md#building-with-docker) (*Recommended*)
122+
- [Install Triton Inference Server without Docker containers](docs/customization_guide/build.md#building-without-docker)
123+
- [Build a custom Triton Inference Server Docker container](docs/customization_guide/compose.md)
124+
- [Build Triton Inference Server from source](docs/customization_guide/build.md#building-on-unsupported-platforms)
125+
- [Build Triton Inference Server for Windows 10](docs/customization_guide/build.md#building-for-windows-10)
126+
- Examples for deploying Triton Inference Server with Kubernetes and Helm on [GCP](deploy/gcp/README.md),
127+
[AWS](deploy/aws/README.md), and [NVIDIA FleetCommand](deploy/fleetcommand/README.md)
128+
129+
### Using Triton
130+
131+
#### Preparing Models for Triton Inference Server
132+
133+
The first step in using Triton to serve your models is to place one or
134+
more models into a [model repository](docs/user_guide/model_repository.md). Depending on
135+
the type of the model and on what Triton capabilities you want to enable for
136+
the model, you may need to create a [model
137+
configuration](docs/user_guide/model_configuration.md) for the model.
138+
139+
- [Add custom operations to Triton if needed by your model](docs/user_guide/custom_operations.md)
140+
- Enable model pipelining with [Model Ensemble](docs/user_guide/architecture.md#ensemble-models)
141+
and [Business Logic Scripting (BLS)](https://github.com/triton-inference-server/python_backend#business-logic-scripting)
142+
- Optimize your models setting [scheduling and batching](docs/user_guide/architecture.md#models-and-schedulers)
143+
parameters and [model instances](docs/user_guide/model_configuration.md#instance-groups).
144+
- Use the [Model Analyzer tool](https://github.com/triton-inference-server/model_analyzer)
145+
to help optimize your model configuration with profiling
146+
- Learn how to [explicitly manage what models are available by loading and
147+
unloading models](docs/user_guide/model_management.md)
148+
149+
#### Configure and Use Triton Inference Server
150+
151+
- Read the [Quick Start Guide](docs/getting_started/quickstart.md) to run Triton Inference
152+
Server on both GPU and CPU
153+
- Triton supports multiple execution engines, called
154+
[backends](https://github.com/triton-inference-server/backend#where-can-i-find-all-the-backends-that-are-available-for-triton), including
155+
[TensorRT](https://github.com/triton-inference-server/tensorrt_backend),
156+
[TensorFlow](https://github.com/triton-inference-server/tensorflow_backend),
157+
[PyTorch](https://github.com/triton-inference-server/pytorch_backend),
158+
[ONNX](https://github.com/triton-inference-server/onnxruntime_backend),
159+
[OpenVINO](https://github.com/triton-inference-server/openvino_backend),
160+
[Python](https://github.com/triton-inference-server/python_backend), and more
161+
- Not all the above backends are supported on every platform supported by Triton.
162+
Look at the
163+
[Backend-Platform Support Matrix](https://github.com/triton-inference-server/backend/blob/main/docs/backend_platform_support_matrix.md)
164+
to learn which backends are supported on your target platform.
165+
- Learn how to [optimize performance](docs/user_guide/optimization.md) using the
166+
[Performance Analyzer](docs/user_guide/perf_analyzer.md) and
167+
[Model Analyzer](https://github.com/triton-inference-server/model_analyzer)
168+
- Learn how to [manage loading and unloading models](docs/user_guide/model_management.md) in
169+
Triton
170+
- Send requests directly to Triton with the [HTTP/REST JSON-based
171+
or gRPC protocols](docs/customization_guide/inference_protocols.md#httprest-and-grpc-protocols)
172+
173+
#### Client Support and Examples
174+
175+
A Triton *client* application sends inference and other requests to Triton. The
176+
[Python and C++ client libraries](https://github.com/triton-inference-server/client)
177+
provide APIs to simplify this communication.
178+
179+
- Review client examples for [C++](https://github.com/triton-inference-server/client/blob/main/src/c%2B%2B/examples),
180+
[Python](https://github.com/triton-inference-server/client/blob/main/src/python/examples),
181+
and [Java](https://github.com/triton-inference-server/client/blob/main/src/java/src/main/java/triton/client/examples)
182+
- Configure [HTTP](https://github.com/triton-inference-server/client#http-options)
183+
and [gRPC](https://github.com/triton-inference-server/client#grpc-options)
184+
client options
185+
- Send input data (e.g. a jpeg image) directly to Triton in the [body of an HTTP
186+
request without any additional metadata](https://github.com/triton-inference-server/server/blob/main/docs/protocol/extension_binary_data.md#raw-binary-request)
187+
188+
### Extend Triton
189+
190+
[Triton Inference Server's architecture](docs/user_guide/architecture.md) is specifically
191+
designed for modularity and flexibility
192+
193+
- [Customize Triton Inference Server container](docs/customization_guide/compose.md) for your use case
194+
- [Create custom backends](https://github.com/triton-inference-server/backend)
195+
in either [C/C++](https://github.com/triton-inference-server/backend/blob/main/README.md#triton-backend-api)
196+
or [Python](https://github.com/triton-inference-server/python_backend)
197+
- Create [decouple backends and models](docs/user_guide/decoupled_models.md) that can send
198+
multiple responses for a request or not send any responses for a request
199+
- Use a [Triton repository agent](docs/customization_guide/repository_agents.md) to add functionality
200+
that operates when a model is loaded and unloaded, such as authentication,
201+
decryption, or conversion
202+
- Deploy Triton on [Jetson and JetPack](docs/user_guide/jetson.md)
203+
- [Use Triton on AWS
204+
Inferentia](https://github.com/triton-inference-server/python_backend/tree/main/inferentia)
205+
206+
### Additional Documentation
207+
208+
- [FAQ](docs/user_guide/faq.md)
209+
- [User Guide](docs/README.md#user-guide)
210+
- [Customization Guide](docs/README.md#customization-guide)
211+
- [Release Notes](https://docs.nvidia.com/deeplearning/triton-inference-server/release-notes/index.html)
212+
- [GPU, Driver, and CUDA Support
213+
Matrix](https://docs.nvidia.com/deeplearning/dgx/support-matrix/index.html)
214+
215+
## Contributing
216+
217+
Contributions to Triton Inference Server are more than welcome. To
218+
contribute please review the [contribution
219+
guidelines](CONTRIBUTING.md). If you have a backend, client,
220+
example or similar contribution that is not modifying the core of
221+
Triton, then you should file a PR in the [contrib
222+
repo](https://github.com/triton-inference-server/contrib).
223+
224+
## Reporting problems, asking questions
225+
226+
We appreciate any feedback, questions or bug reporting regarding this project.
227+
When posting [issues in GitHub](https://github.com/triton-inference-server/server/issues),
228+
follow the process outlined in the [Stack Overflow document](https://stackoverflow.com/help/mcve).
229+
Ensure posted examples are:
230+
- minimal – use as little code as possible that still produces the
231+
same problem
232+
- complete – provide all parts needed to reproduce the problem. Check
233+
if you can strip external dependencies and still show the problem. The
234+
less time we spend on reproducing problems the more time we have to
235+
fix it
236+
- verifiable – test the code you're about to provide to make sure it
237+
reproduces the problem. Remove all other problems that are not
238+
related to your request/question.
239+
240+
## For more information
241+
242+
Please refer to the [NVIDIA Developer Triton page](https://developer.nvidia.com/nvidia-triton-inference-server)
243+
for more information.

‎RELEASE.md

+91
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,91 @@
1+
<!--
2+
# Copyright 2018-2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3+
#
4+
# Redistribution and use in source and binary forms, with or without
5+
# modification, are permitted provided that the following conditions
6+
# are met:
7+
# * Redistributions of source code must retain the above copyright
8+
# notice, this list of conditions and the following disclaimer.
9+
# * Redistributions in binary form must reproduce the above copyright
10+
# notice, this list of conditions and the following disclaimer in the
11+
# documentation and/or other materials provided with the distribution.
12+
# * Neither the name of NVIDIA CORPORATION nor the names of its
13+
# contributors may be used to endorse or promote products derived
14+
# from this software without specific prior written permission.
15+
#
16+
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
17+
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18+
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19+
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
20+
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
21+
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22+
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23+
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
24+
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25+
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26+
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27+
-->
28+
29+
# Release Notes for 2.29.0
30+
31+
## New Freatures and Improvements
32+
33+
* Improvements to container and non-container builds on Windows.
34+
35+
* Refer to the 22.12 column of the
36+
[Frameworks Support Matrix](https://docs.nvidia.com/deeplearning/frameworks/support-matrix/index.html)
37+
for container image versions on which the 22.12 inference server container is
38+
based.
39+
40+
## Known Issues
41+
42+
* In some rare cases Triton might overwrite input tensors while they are still
43+
in use which leads to corrupt input data being used for inference with
44+
TensorRT models. If you encounter accuracy issues with your TensorRT model,
45+
you can work-around the issue by
46+
[enabling the output_copy_stream option](https://github.com/triton-inference-server/common/blob/r22.12/protobuf/model_config.proto#L843-L852)
47+
in your model's configuration.
48+
49+
* Some systems which implement `malloc()` may not release memory back to the
50+
operating system right away causing a false memory leak. This can be mitigated
51+
by using a different malloc implementation. Tcmalloc is installed in the
52+
Triton container and can be
53+
[used by specifying the library in LD_PRELOAD](https://github.com/triton-inference-server/server/blob/r22.12/docs/user_guide/model_management.md#model-control-mode-explicit).
54+
55+
* When using a custom operator for the PyTorch backend, the operator may not be
56+
loaded due to undefined Python library symbols. This can be work-around by
57+
[specifying Python library in LD_PRELOAD](https://github.com/triton-inference-server/server/blob/r22.12/qa/L0_custom_ops/test.sh#L114-L117).
58+
59+
* Auto-complete may cause an increase in server start time. To avoid a start
60+
time increase, users can provide the full model configuration and launch the
61+
server with `--disable-auto-complete-config`.
62+
63+
* Auto-complete does not support PyTorch models due to lack of metadata in the
64+
model. It can only verify that the number of inputs and the input names
65+
matches what is specified in the model configuration. There is no model
66+
metadata about the number of outputs and datatypes. Related PyTorch bug:
67+
https://github.com/pytorch/pytorch/issues/38273
68+
69+
* Perf Analyzer stability criteria has been changed which may result in
70+
reporting instability for scenarios that were previously considered stable.
71+
This change has been made to improve the accuracy of Perf Analyzer results.
72+
If you observe this message, it can be resolved by increasing the
73+
`--measurement-interval` in the time windows mode or
74+
`--measurement-request-count` in the count windows mode.
75+
76+
* Triton Client PIP wheels for ARM SBSA are not available from PyPI and pip will
77+
install an incorrect Jetson version of Triton Client library for Arm SBSA.
78+
79+
The correct client wheel file can be pulled directly from the Arm SBSA SDK
80+
image and manually installed.
81+
82+
* Traced models in PyTorch seem to create overflows when int8 tensor values are
83+
transformed to int32 on the GPU.
84+
85+
Refer to https://github.com/pytorch/pytorch/issues/66930 for more information.
86+
87+
* Triton cannot retrieve GPU metrics with MIG-enabled GPU devices (A100 and A30).
88+
89+
* Triton metrics might not work if the host machine is running a separate DCGM
90+
agent on bare-metal or in a container.
91+

0 commit comments

Comments
 (0)