Get stuck on the stage "Engine version 1.0.0 found in the config file, assuming engine(s) built by new builder API."

[ZCJ1111]

Description
When building a TensorRT-LLM engine for Qwen3-235B (MoE) in a rootless Podman container, the build process completes the engine serialization but then hangs indefinitely at the log message:
[TensorRT-LLM][INFO] Engine version 1.0.0 found in the config file, assuming engine(s) built by new builder API. Not sure why this happened?

Triton Information
I am using the image nvcr.io/nvidia/tritonserver:25.10-trtllm-python-py3 and run it by podman in a rootless mode.

To Reproduce

podman run --rm -it \
  --gpus all \
  --security-opt label=disable \
  --network host \
  --shm-size=64g \
  --pids-limit=-1 \
  --ulimit memlock=-1 \
  --ulimit stack=67108864 \
    nvcr.io/nvidia/tritonserver:25.10-trtllm-python-py3 bash \

python3 examples/qwen/convert_checkpoint.py \
    --model_dir /models/source \
    --output_dir /models/engines/ckpt \
    --dtype auto\
    --tp_size 2 \
    --pp_size 2 \
    --moe_tp_size 2 \
    --workers 4

trtllm-build  --checkpoint_dir /models/engines/ckpt \
    --output_dir /models/engines/final \
    --gemm_plugin auto \
    --max_batch_size 256 \
    --max_input_len 2048 \
    --max_seq_len 8192 \
    --max_num_tokens 16384 \
    --workers 4

python3 /app/scripts/launch_triton_server.py --world_size=4 --model_repo=/models

Expected behavior

After running the launch_triton_server.py, it should appear the model loaded, i.e. preprocessing, tensorrtllm, postprocessing and listening port, 8000, 8001, 8002. However it stuck at [TensorRT-LLM][INFO] Engine version 1.0.0 found in the config file, assuming engine(s) built by new builder API.

[ZCJ1111]

NVIDIA/TensorRT-LLM#2305 (comment) does not help, I turned off the CPU virtulisation feature, but still stuck.