The engine-build multimodal workflow that used to live here (build_multimodal_engine.py / run.py / eval.py on top of trtllm-build) was removed together with the legacy TensorRT backend.

Multimodal models are supported on the PyTorch backend. See:

• Supported models
• trtllm-serve and the LLM Python API
• Multimodal serving examples under examples/llm-api/ and examples/serve/

qwen_image_bench_eval.py runs the Qwen-Image-Bench VLM judge through TensorRT-LLM to evaluate an image generated by another image or vision-language model. Provide the original generation prompt and the generated image path:

python examples/models/core/multimodal/qwen_image_bench_eval.py \
    --model_path /path/to/Qwen-Image-Bench \
    --prompt "A cute cat playing piano" \
    --image_path examples/visual_gen/cat_piano.png \
    --output_path qwen_image_bench_result.json \
    --include_raw_outputs

The script evaluates all five Qwen-Image-Bench level-1 dimensions by default: Quality, Aesthetics, Alignment, Real-world Fidelity, and Creative Generation.