trtexec error on Thor
#1
by Orangecx - opened
NVIDIA Jetson AGX Thor with TensorRT 10.13.3 used
Segmentation fault (core dumped) occured when using command:
/usr/src/tensorrt/bin/trtexec --onnx=./embedl/embedl_sam3_quant.onnx \
--fp16 --int8 \
--builderOptimizationLevel=2 \
--memPoolSize=workspace:4294967296 \
--timingCacheFile=embedl_sam3_timing_cache.bin \
--saveEngine=embedl_sam3_quant.engine
log is
[05/25/2026-11:42:29] [W] [TRT] /quant_469/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_473/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_470/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_474/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_471/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_475/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_479/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_477/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_476/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_480/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_478/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_481/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_482/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_486/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_483/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_487/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_484/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_488/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_485/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_489/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_493/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_491/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_490/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_494/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_492/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_495/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_496/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_497/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_503/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_512/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_498/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_504/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_513/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_499/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_505/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_514/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:29] [W] [TRT] /quant_506/DequantizeLinear [SCALE] has invalid precision Int8, ignored.
[05/25/2026-11:42:30] [I] [TRT] Global timing cache in use. Profiling results in this builder pass will be stored.
[05/25/2026-11:42:39] [I] [TRT] Compiler backend is used during engine build.
Segmentation fault (core dumped)
Hi,
Thanks for opening the issue. We haven't still added fp8 and Blackwell support for this model. The quantization and QDQ patterns are only for int8 quantization.
See the full tutorial to see how the quantization was performed: https://docs.embedl.com/embedl-deploy/latest/auto_tutorials/sam3.html
We are working on adding support for Blackwell GPUs.