You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@@ -12,15 +12,15 @@ This backend was developed using the existing [Triton PyTorch Backend](https://g
12
12
13
13
This backend is only currently available for **linux arm64** platforms.
14
14
15
-
## Build the TFLite Backend
16
-
The TFLite backend can be built either integrated with the build process for the [triton server repo](https://github.com/triton-inference-server/server) or it may be built independently using only this repository.
15
+
## Build the ArmNN TFLite Backend
16
+
The ArmNN TFLite backend can be built either integrated with the build process for the [triton server repo](https://github.com/triton-inference-server/server) or it may be built independently using only this repository.
17
17
18
18
### Build with Triton Build Convenience Script
19
-
The easiest way to get up and running with the triton tflite backend is to build a custom triton docker image using the `build.py` script available in the triton server repo.
19
+
The easiest way to get up and running with the triton armnn tflite backend is to build a custom triton docker image using the `build.py` script available in the triton server repo.
20
20
21
-
To build a triton server docker image with the tflite backend built in simply run the following command from the root of the server repo:
21
+
To build a triton server docker image with the armnn tflite backend built in simply run the following command from the root of the server repo:
Note that for MALI GPU tuning level the value corresponds to the following: `(0=UseOnly(default) | 1=RapidTuning | 2=NormalTuning | 3=ExhaustiveTuning)`
150
148
151
149
### XNNPACK Delegate Optimization Options
152
150
Users also have the ability to specify XNNPACK specific optimizations.
The best way to run the TFLite backend on a platform with a MALI GPU is via Docker. For example on a hikey 970, we can run the following after building our custom tritonserver image using the command from the build with convenience script above:
160
+
## Running ArmNN TFLite Backend on MALI GPU
161
+
The best way to run the ArmNN TFLite backend on a platform with a MALI GPU is via Docker. For example on a hikey 970, we can run the following after building our custom tritonserver image using the command from the build with convenience script above:
164
162
```
165
163
docker run --rm -it --device /dev/mali0 -v /usr/lib/aarch64-linux-gnu/libmali.so:/usr/lib/aarch64-linux-gnu/libmali.so -v <full path to your model repo on host>:/models -p 8000:8000 -p 8001:8001 -p 8002:8002 tritonserver:latest
0 commit comments