difference between .trt, .etlt, .trt.int8 and etlt_b1_gpu0_int8.engine files in TLT. Is configuring each file in Deepstream different from each other.
In deep stream config file, there are three different path parameters
- tlt-model-key
- GPU engine file
- DLA engine file
What is the purpose of each one.
- The .trt file is trt engine normally. But this ext is named by end user. You can save a trt engine to any name you like. But usually we set to .trt or .engine
- The .etlt file is an encrypted TLT file. During model export TLT model is encrypted with a private key. See https://pgambrill.gitlab-master-pages.nvidia.com/tlt-docs/text/exporting_model.html?highlight=etlt#
- The .trt.int8 file is normally an int8 trt engine.
- The etlt_b1_gpu0_int8.engine is generated by deepstream app. During its inference, it will save a trt engine.
For tlt-model-key, please see https://pgambrill.gitlab-master-pages.nvidia.com/tlt-docs/text/deploying_to_deepstream.html#integrating-a-detectnet-v2-model
More, can you give more details about “GPU engine file” and " DLA engine file"? Seems that there are not these parameters in DS.
1 Like