Skip to content

Commit 386d05d

Browse files
mc-nvmc-nv
committed
Update README for 22.10 release
1 parent 549ab86 commit 386d05d

File tree

2 files changed

+308
-2
lines changed

2 files changed

+308
-2
lines changed

‎README.md

+214-2
Original file line numberDiff line numberDiff line change
@@ -28,5 +28,217 @@
2828

2929
# Triton Inference Server
3030

31-
**NOTE: You are currently on the r21.03 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+
----
34+
Triton Inference Server is an open source inference serving software that
35+
streamlines AI inferencing. Triton enables teams to deploy any AI model from
36+
multiple deep learning and machine learning frameworks, including TensorRT,
37+
TensorFlow, PyTorch, ONNX, OpenVINO, Python, RAPIDS FIL, and more. Triton
38+
supports inference across cloud, data center,edge and embedded devices on NVIDIA
39+
GPUs, x86 and ARM CPU, or AWS Inferentia. Triton delivers optimized performance
40+
for many query types, including real time, batched, ensembles and audio/video
41+
streaming.
42+
43+
Major features include:
44+
45+
- [Supports multiple deep learning
46+
frameworks](https://github.com/triton-inference-server/backend#where-can-i-find-all-the-backends-that-are-available-for-triton)
47+
- [Supports multiple machine learning
48+
frameworks](https://github.com/triton-inference-server/fil_backend)
49+
- [Concurrent model
50+
execution](docs/user_guide/architecture.md#concurrent-model-execution)
51+
- [Dynamic batching](docs/user_guide/model_configuration.md#dynamic-batcher)
52+
- [Sequence batching](docs/user_guide/model_configuration.md#sequence-batcher) and
53+
[implicit state management](docs/user_guide/architecture.md#implicit-state-management)
54+
for stateful models
55+
- Provides [Backend API](https://github.com/triton-inference-server/backend) that
56+
allows adding custom backends and pre/post processing operations
57+
- Model pipelines using
58+
[Ensembling](docs/user_guide/architecture.md#ensemble-models) or [Business
59+
Logic Scripting
60+
(BLS)](https://github.com/triton-inference-server/python_backend#business-logic-scripting)
61+
- [HTTP/REST and GRPC inference
62+
protocols](docs/customization_guide/inference_protocols.md) based on the community
63+
developed [KServe
64+
protocol](https://github.com/kserve/kserve/tree/master/docs/predict-api/v2)
65+
- A [C API](docs/customization_guide/inference_protocols.md#in-process-triton-server-api) and
66+
[Java API](docs/customization_guide/inference_protocols.md#java-bindings-for-in-process-triton-server-api)
67+
allow Triton to link directly into your application for edge and other in-process use cases
68+
- [Metrics](docs/user_guide/metrics.md) indicating GPU utilization, server
69+
throughput, server latency, and more
70+
71+
Join the [Triton and TensorRT community](https://www.nvidia.com/en-us/deep-learning-ai/triton-tensorrt-newsletter/) and
72+
stay current on the latest product updates, bug fixes, content, best practices,
73+
and more. Need enterprise support? NVIDIA global support is available for Triton
74+
Inference Server with the
75+
[NVIDIA AI Enterprise software suite](https://www.nvidia.com/en-us/data-center/products/ai-enterprise/).
76+
77+
## Serve a Model in 3 Easy Steps
78+
79+
```bash
80+
# Step 1: Create the example model repository
81+
git clone -b r22.10 https://github.com/triton-inference-server/server.git
82+
cd server/docs/examples
83+
./fetch_models.sh
84+
85+
# Step 2: Launch triton from the NGC Triton container
86+
docker run --gpus=1 --rm --net=host -v ${PWD}/model_repository:/models nvcr.io/nvidia/tritonserver:22.10-py3 tritonserver --model-repository=/models
87+
88+
# Step 3: Sending an Inference Request
89+
# In a separate console, launch the image_client example from the NGC Triton SDK container
90+
docker run -it --rm --net=host nvcr.io/nvidia/tritonserver:22.10-py3-sdk
91+
/workspace/install/bin/image_client -m densenet_onnx -c 3 -s INCEPTION /workspace/images/mug.jpg
92+
93+
# Inference should return the following
94+
Image '/workspace/images/mug.jpg':
95+
15.346230 (504) = COFFEE MUG
96+
13.224326 (968) = CUP
97+
10.422965 (505) = COFFEEPOT
98+
```
99+
Please read the [QuickStart](docs/getting_started/quickstart.md) guide for additional information
100+
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).
101+
102+
## Examples and Tutorials
103+
104+
Check out [NVIDIA LaunchPad](https://www.nvidia.com/en-us/data-center/products/ai-enterprise-suite/trial/)
105+
for free access to a set of hands-on labs with Triton Inference Server hosted on
106+
NVIDIA infrastructure.
107+
108+
Specific end-to-end examples for popular models, such as ResNet, BERT, and DLRM
109+
are located in the
110+
[NVIDIA Deep Learning Examples](https://github.com/NVIDIA/DeepLearningExamples)
111+
page on GitHub. The
112+
[NVIDIA Developer Zone](https://developer.nvidia.com/nvidia-triton-inference-server)
113+
contains additional documentation, presentations, and examples.
114+
115+
## Documentation
116+
117+
### Build and Deploy
118+
119+
The recommended way to build and use Triton Inference Server is with Docker
120+
images.
121+
122+
- [Install Triton Inference Server with Docker containers](docs/customization_guide/build.md#building-with-docker) (*Recommended*)
123+
- [Install Triton Inference Server without Docker containers](docs/customization_guide/build.md#building-without-docker)
124+
- [Build a custom Triton Inference Server Docker container](docs/customization_guide/compose.md)
125+
- [Build Triton Inference Server from source](docs/customization_guide/build.md#building-on-unsupported-platforms)
126+
- [Build Triton Inference Server for Windows 10](docs/customization_guide/build.md#building-for-windows-10)
127+
- Examples for deploying Triton Inference Server with Kubernetes and Helm on [GCP](deploy/gcp/README.md),
128+
[AWS](deploy/aws/README.md), and [NVIDIA FleetCommand](deploy/fleetcommand/README.md)
129+
130+
### Using Triton
131+
132+
#### Preparing Models for Triton Inference Server
133+
134+
The first step in using Triton to serve your models is to place one or
135+
more models into a [model repository](docs/user_guide/model_repository.md). Depending on
136+
the type of the model and on what Triton capabilities you want to enable for
137+
the model, you may need to create a [model
138+
configuration](docs/user_guide/model_configuration.md) for the model.
139+
140+
- [Add custom operations to Triton if needed by your model](docs/user_guide/custom_operations.md)
141+
- Enable model pipelining with [Model Ensemble](docs/user_guide/architecture.md#ensemble-models)
142+
and [Business Logic Scripting (BLS)](https://github.com/triton-inference-server/python_backend#business-logic-scripting)
143+
- Optimize your models setting [scheduling and batching](docs/user_guide/architecture.md#models-and-schedulers)
144+
parameters and [model instances](docs/user_guide/model_configuration.md#instance-groups).
145+
- Use the [Model Analyzer tool](https://github.com/triton-inference-server/model_analyzer)
146+
to help optimize your model configuration with profiling
147+
- Learn how to [explicitly manage what models are available by loading and
148+
unloading models](docs/user_guide/model_management.md)
149+
150+
#### Configure and Use Triton Inference Server
151+
152+
- Read the [Quick Start Guide](docs/getting_started/quickstart.md) to run Triton Inference
153+
Server on both GPU and CPU
154+
- Triton supports multiple execution engines, called
155+
[backends](https://github.com/triton-inference-server/backend#where-can-i-find-all-the-backends-that-are-available-for-triton), including
156+
[TensorRT](https://github.com/triton-inference-server/tensorrt_backend),
157+
[TensorFlow](https://github.com/triton-inference-server/tensorflow_backend),
158+
[PyTorch](https://github.com/triton-inference-server/pytorch_backend),
159+
[ONNX](https://github.com/triton-inference-server/onnxruntime_backend),
160+
[OpenVINO](https://github.com/triton-inference-server/openvino_backend),
161+
[Python](https://github.com/triton-inference-server/python_backend), and more
162+
- Not all the above backends are supported on every platform supported by Triton.
163+
Look at the
164+
[Backend-Platform Support Matrix](https://github.com/triton-inference-server/backend/blob/main/docs/backend_platform_support_matrix.md)
165+
to learn which backends are supported on your target platform.
166+
- Learn how to [optimize performance](docs/user_guide/optimization.md) using the
167+
[Performance Analyzer](docs/user_guide/perf_analyzer.md) and
168+
[Model Analyzer](https://github.com/triton-inference-server/model_analyzer)
169+
- Learn how to [manage loading and unloading models](docs/user_guide/model_management.md) in
170+
Triton
171+
- Send requests directly to Triton with the [HTTP/REST JSON-based
172+
or gRPC protocols](docs/customization_guide/inference_protocols.md#httprest-and-grpc-protocols)
173+
174+
#### Client Support and Examples
175+
176+
A Triton *client* application sends inference and other requests to Triton. The
177+
[Python and C++ client libraries](https://github.com/triton-inference-server/client)
178+
provide APIs to simplify this communication.
179+
180+
- Review client examples for [C++](https://github.com/triton-inference-server/client/blob/main/src/c%2B%2B/examples),
181+
[Python](https://github.com/triton-inference-server/client/blob/main/src/python/examples),
182+
and [Java](https://github.com/triton-inference-server/client/blob/main/src/java/src/main/java/triton/client/examples)
183+
- Configure [HTTP](https://github.com/triton-inference-server/client#http-options)
184+
and [gRPC](https://github.com/triton-inference-server/client#grpc-options)
185+
client options
186+
- Send input data (e.g. a jpeg image) directly to Triton in the [body of an HTTP
187+
request without any additional metadata](https://github.com/triton-inference-server/server/blob/main/docs/protocol/extension_binary_data.md#raw-binary-request)
188+
189+
### Extend Triton
190+
191+
[Triton Inference Server's architecture](docs/user_guide/architecture.md) is specifically
192+
designed for modularity and flexibility
193+
194+
- [Customize Triton Inference Server container](docs/customization_guide/compose.md) for your use case
195+
- [Create custom backends](https://github.com/triton-inference-server/backend)
196+
in either [C/C++](https://github.com/triton-inference-server/backend/blob/main/README.md#triton-backend-api)
197+
or [Python](https://github.com/triton-inference-server/python_backend)
198+
- Create [decouple backends and models](docs/user_guide/decoupled_models.md) that can send
199+
multiple responses for a request or not send any responses for a request
200+
- Use a [Triton repository agent](docs/customization_guide/repository_agents.md) to add functionality
201+
that operates when a model is loaded and unloaded, such as authentication,
202+
decryption, or conversion
203+
- Deploy Triton on [Jetson and JetPack](docs/user_guide/jetson.md)
204+
- [Use Triton on AWS
205+
Inferentia](https://github.com/triton-inference-server/python_backend/tree/main/inferentia)
206+
207+
### Additional Documentation
208+
209+
- [FAQ](docs/user_guide/faq.md)
210+
- [User Guide](docs/README.md#user-guide)
211+
- [Customization Guide](docs/README.md#customization-guide)
212+
- [Release Notes](https://docs.nvidia.com/deeplearning/triton-inference-server/release-notes/index.html)
213+
- [GPU, Driver, and CUDA Support
214+
Matrix](https://docs.nvidia.com/deeplearning/dgx/support-matrix/index.html)
215+
216+
## Contributing
217+
218+
Contributions to Triton Inference Server are more than welcome. To
219+
contribute please review the [contribution
220+
guidelines](CONTRIBUTING.md). If you have a backend, client,
221+
example or similar contribution that is not modifying the core of
222+
Triton, then you should file a PR in the [contrib
223+
repo](https://github.com/triton-inference-server/contrib).
224+
225+
## Reporting problems, asking questions
226+
227+
We appreciate any feedback, questions or bug reporting regarding this project.
228+
When posting [issues in GitHub](https://github.com/triton-inference-server/server/issues),
229+
follow the process outlined in the [Stack Overflow document](https://stackoverflow.com/help/mcve).
230+
Ensure posted examples are:
231+
- minimal – use as little code as possible that still produces the
232+
same problem
233+
- complete – provide all parts needed to reproduce the problem. Check
234+
if you can strip external dependencies and still show the problem. The
235+
less time we spend on reproducing problems the more time we have to
236+
fix it
237+
- verifiable – test the code you're about to provide to make sure it
238+
reproduces the problem. Remove all other problems that are not
239+
related to your request/question.
240+
241+
## For more information
242+
243+
Please refer to the [NVIDIA Developer Triton page](https://developer.nvidia.com/nvidia-triton-inference-server)
244+
for more information.

‎RELEASE.md

+94
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,94 @@
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 2.27.0
30+
31+
## New Freatures and Improvements
32+
33+
* Added an [example](https://github.com/triton-inference-server/python_backend/tree/r22.10/examples/jax)
34+
to demonstrate the use of JAX in python models.
35+
36+
* Improved and enhanced [Server Wrapper API](https://github.com/triton-inference-server/developer_tools/tree/r22.10/server)
37+
to include missing features such as
38+
[decoupled model](https://github.com/triton-inference-server/server/blob/r22.10/docs/user_guide/decoupled_models.md)
39+
and tracing support.
40+
41+
* Multiple concurrent models can be profiled and analyzed by Model Analyzer. See
42+
[Multi-Model Search Mode](https://github.com/triton-inference-server/model_analyzer/blob/main/docs/config_search.md#multi-model-search-mode)
43+
for additional details.
44+
45+
46+
47+
## Known Issues
48+
49+
* Triton's TensorRT support depends on the CUDA event synchronization. In some
50+
rare cases the events may be triggered earlier than expected, causing Triton
51+
to overwrite input tensors while they are still in use and leading to corrupt
52+
input data being used for inference. If you encounter accuracy issues with
53+
your TensorRT model, you can work-around the issue by
54+
[enabling the output_copy_stream option](https://github.com/triton-inference-server/common/blob/r22.09/protobuf/model_config.proto#L843-L852)
55+
in your model's configuration.
56+
57+
* When using a custom operator for the PyTorch backend, the operator may not be
58+
loaded due to undefined Python library symbols. This can be work-around by
59+
[specifying Python library in LD_PRELOAD](https://github.com/triton-inference-server/server/blob/r22.09/qa/L0_custom_ops/test.sh#L114-L117).
60+
61+
* Auto-complete may cause an increase in server start time. To avoid a start
62+
time increase, users can provide the full model configuration and launch the
63+
server with `--disable-auto-complete-config`.
64+
65+
* Auto-complete does not support PyTorch models due to lack of metadata in the
66+
model. It can only verify that the number of inputs and the input names
67+
matches what is specified in the model configuration. There is no model
68+
metadata about the number of outputs and datatypes. Related PyTorch bug:
69+
https://github.com/pytorch/pytorch/issues/38273
70+
71+
* Perf Analyzer stability criteria has been changed which may result in
72+
reporting instability for scenarios that were previously considered stable.
73+
This change has been made to improve the accuracy of Perf Analyzer results.
74+
If you observe this message, it can be resolved by increasing the
75+
`--measurement-interval` in the time windows mode or
76+
`--measurement-request-count` in the count windows mode.
77+
78+
* Triton Client PIP wheels for ARM SBSA are not available from PyPI and pip will
79+
install an incorrect Jetson version of Triton Client library for Arm SBSA.
80+
81+
The correct client wheel file can be pulled directly from the Arm SBSA SDK
82+
image and manually installed.
83+
84+
* Traced models in PyTorch seem to create overflows when int8 tensor values are
85+
transformed to int32 on the GPU.
86+
87+
Refer to https://github.com/pytorch/pytorch/issues/66930 for more information.
88+
89+
* Triton cannot retrieve GPU metrics with MIG-enabled GPU devices (A100 and A30).
90+
91+
* Triton metrics might not work if the host machine is running a separate DCGM
92+
agent on bare-metal or in a container.
93+
94+

0 commit comments

Comments
 (0)