[Model Requests] Support Qwen2.5-VL Architecture

[mtezgider]

I would like to request support for the Qwen2.5-VL architecture in TensorRT-LLM.

When I try to convert the model to an LLM checkpoint and build the engine, I encounter the following error:

Traceback (most recent call last): File "/usr/local/bin/trtllm-build", line 8, in <module> sys.exit(main()) ^^^^^^ File "/usr/local/lib/python3.12/dist-packages/tensorrt_llm/commands/build.py", line 517, in main model_config = PretrainedConfig.from_json_file(config_path) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/tensorrt_llm/models/modeling_utils.py", line 434, in from_json_file obj = cls.from_dict(config) ^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/tensorrt_llm/models/modeling_utils.py", line 416, in from_dict model_cls = MODEL_MAP[config['architecture']] ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^ KeyError: 'Qwen2_5_VLForConditionalGeneration'

It seems that Qwen2.5-VL is not yet included in the MODEL_MAP within TensorRT-LLM. Since Qwen2 VL models are already supported, adding Qwen2.5-VL would be a valuable extension, particularly for multimodal tasks (text + image understanding).

Request:
Could you provide insights on whether there are plans to support this architecture in future releases?

Thanks!