NVIDIA gptj - "Could not locate GPT-J fp8 quantized checkpoint model path"

[saibulusu]

Using a system with NVIDIA Ampere GPUs.
system ID: KnownSystem.A100_SXM4_40GBx8

After downloading model, downloading data, and preprocessing data, I do not see the model listed under the GPTJ-6B folder.

Running make generate_engines RUN_ARGS="--benchmarks=gptj --scenarios=Offline --config_ver=default --test_mode=AccuracyOnly" in my docker container.

For reference, both bert & 3d-unet benchmarks are working.

[saibulusu]

When I run the command 'rm -rf build/TRTLLM && make clone_trt_llm && make build_trt_llm' to prepare for quantization, I am seeing this error below:

The pytorch version seems correct

I am using CUDA 12.4