TensorRT - Revision history

SanzSkuy at 02:07, 7 November 2025

2025-11-07T02:07:47Z

← Older revision		Revision as of 02:07, 7 November 2025
Line 6:		Line 6:
	cd TensorRT-RTX-1.1.1.26		cd TensorRT-RTX-1.1.1.26
	export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc		export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc
	ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec		sudo ln -s $(realpath bin/tensorrt_rtx) /usr/local/bin/trtexec
	</syntaxhighlight>		</syntaxhighlight>
	Convert		Convert

SanzSkuy: /* Triton Server */

2025-09-27T02:35:01Z

Triton Server

@@ Line 177: / Line 177: @@
    max_queue_delay_microseconds: 5000
+}
-</syntaxhighlight>
+{| class="wikitable"

SanzSkuy at 01:48, 27 September 2025

2025-09-27T01:48:14Z

@@ Line 1: / Line 1: @@
 High performance inference engine. <br>
 pertama-tama download <code>tar.gz</code>-nya di https://developer.nvidia.com/tensorrt-rtx
@@ Line 58: / Line 59: @@
          --exportOutput=test_result.json \
          --verbose
 </syntaxhighlight>

SanzSkuy at 13:41, 26 September 2025

2025-09-26T13:41:45Z

← Older revision		Revision as of 13:41, 26 September 2025
Line 6:		Line 6:
	export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc		export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc
	ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec		ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec
	<syntaxhighlight>		</syntaxhighlight>
			Convert
			<syntaxhighlight lang=bash>
	# Konversi dasar ONNX ke TensorRT		# Konversi dasar ONNX ke TensorRT
	trtexec --onnx=mobilenetv2-7.onnx --saveEngine=mobilenetv2.trt		trtexec --onnx=mobilenetv2-7.onnx --saveEngine=mobilenetv2.trt
Line 56:		Line 58:
	--exportOutput=test_result.json \		--exportOutput=test_result.json \
	--verbose		--verbose
	<syntaxhighlight ~~lang=bash~~>		</syntaxhighlight>

SanzSkuy at 13:40, 26 September 2025

2025-09-26T13:40:58Z

← Older revision		Revision as of 13:40, 26 September 2025
Line 7:		Line 7:
	ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec		ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec
	<syntaxhighlight>		<syntaxhighlight>
	~~Convert~~
	~~<syntaxhighlight lang=bash>~~
	# Konversi dasar ONNX ke TensorRT		# Konversi dasar ONNX ke TensorRT
	trtexec --onnx=mobilenetv2-7.onnx --saveEngine=mobilenetv2.trt		trtexec --onnx=mobilenetv2-7.onnx --saveEngine=mobilenetv2.trt
Line 58:		Line 56:
	--exportOutput=test_result.json \		--exportOutput=test_result.json \
	--verbose		--verbose
	<syntaxhighlight>		<syntaxhighlight lang=bash>

SanzSkuy: Created page with "High performance inference engine.
pertama-tama download `tar.gz`-nya di https://developer.nvidia.com/tensorrt-rtx tar -xf TensorRT-RTX-1.1.1.26.Linux.x86_64-gnu.cuda-12.9.tar.gz cd TensorRT-RTX-1.1.1.26 export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec Convert # Konversi dasar ONNX ke TensorRT trtex..."

2025-09-26T13:40:20Z

Created page with "High performance inference engine. <br> pertama-tama download <code>tar.gz</code>-nya di https://developer.nvidia.com/tensorrt-rtx <syntaxhighlight lang=bash> tar -xf TensorRT-RTX-1.1.1.26.Linux.x86_64-gnu.cuda-12.9.tar.gz cd TensorRT-RTX-1.1.1.26 export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec <syntaxhighlight> Convert <syntaxhighlight lang=bash> # Konversi dasar ONNX ke TensorRT trtex..."

New page

High performance inference engine. <br>
pertama-tama download <code>tar.gz</code>-nya di https://developer.nvidia.com/tensorrt-rtx
<syntaxhighlight lang=bash>
tar -xf TensorRT-RTX-1.1.1.26.Linux.x86_64-gnu.cuda-12.9.tar.gz
cd TensorRT-RTX-1.1.1.26
export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH # masukin .bashrc atau .zshrc
ln -s $(realpath tensorrt_rtx) /usr/local/bin/trtexec
<syntaxhighlight>
Convert
<syntaxhighlight lang=bash>
# Konversi dasar ONNX ke TensorRT
trtexec --onnx=mobilenetv2-7.onnx --saveEngine=mobilenetv2.trt

# Konversi dengan optimasi tambahan (direkomendasikan)
trtexec --onnx=mobilenetv2-7.onnx \
--saveEngine=mobilenetv2_optimized.trt \
--minShapes=data:1x3x224x224 \
--optShapes=data:1x3x224x224 \
--maxShapes=data:1x3x224x224 \
--builderOptimizationLevel=5 \
--avgTiming=16 \
--memPoolSize=workspace:2G \
--verbose

# Konversi dengan dynamic shapes (jika perlu batch size fleksibel)
trtexec --onnx=mobilenetv2-7.onnx \
--saveEngine=mobilenetv2_dynamic.trt \
--minShapes=data:1x3x224x224 \
--optShapes=data:4x3x224x224 \
--maxShapes=data:16x3x224x224 \
--builderOptimizationLevel=5 \
--verbose

# Konversi dengan precision FP16 untuk performa lebih cepat (jika GPU mendukung)
trtexec --onnx=mobilenetv2-7.onnx \
--saveEngine=mobilenetv2_fp16.trt \
--minShapes=data:1x3x224x224 \
--optShapes=data:1x3x224x224 \
--maxShapes=data:1x3x224x224 \
--fp16 \
--builderOptimizationLevel=5 \
--verbose

# Konversi dengan hardware compatibility (untuk portabilitas GPU)
trtexec --onnx=mobilenetv2-7.onnx \
--saveEngine=mobilenetv2_compatible.trt \
--minShapes=data:1x3x224x224 \
--optShapes=data:1x3x224x224 \
--maxShapes=data:1x3x224x224 \
--hardwareCompatibilityLevel=ampere+ \
--verbose

# Test inference setelah konversi
trtexec --loadEngine=mobilenet.trt \
--shapes=data:1x3x224x224 \
--iterations=100 \
--avgRuns=10 \
--exportOutput=test_result.json \
--verbose
<syntaxhighlight>