Simplify tensorrt building (#13851)

* Simplify tensorrt building

* Use engine context cache
This commit is contained in:
Nicolas Mowen 2024-09-20 09:49:44 -06:00 committed by GitHub
parent 535bf6e4b9
commit 1a51ce712c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 3 additions and 3 deletions

View File

@ -3,8 +3,6 @@
# https://askubuntu.com/questions/972516/debian-frontend-environment-variable
ARG DEBIAN_FRONTEND=noninteractive
ARG TRT_BASE=nvcr.io/nvidia/tensorrt:23.03-py3
# Make this a separate target so it can be built/cached optionally
FROM wheels as trt-wheels
ARG DEBIAN_FRONTEND
@ -15,7 +13,7 @@ COPY docker/tensorrt/requirements-amd64.txt /requirements-tensorrt.txt
RUN mkdir -p /trt-wheels && pip3 wheel --wheel-dir=/trt-wheels -r /requirements-tensorrt.txt
# Build CuDNN
FROM ${TRT_BASE} AS cudnn-deps
FROM wget AS cudnn-deps
ARG COMPUTE_LEVEL

View File

@ -49,7 +49,9 @@ class ONNXDetector(DetectionApi):
"trt_timing_cache_enable": True,
"trt_timing_cache_path": "/config/model_cache/tensorrt/ort",
"trt_engine_cache_enable": True,
"trt_dump_ep_context_model": True,
"trt_engine_cache_path": "/config/model_cache/tensorrt/ort/trt-engines",
"trt_ep_context_file_path": "/config/model_cache/tensorrt/ort",
}
)
elif provider == "OpenVINOExecutionProvider":