[Bug] MLCEngine model parameter gets converted to a tuple internally but not handled correctly

[hank]

🐛 Bug

MLCEngine model parameter gets converted to a tuple internally but not handled correctly.

To Reproduce

Steps to reproduce the behavior:

1. Create a script as shown in the code sample
2. Run the script

from mlc_llm import MLCEngine

# Create engine
model = "./dist/prebuilt/Llama-3-8B-Instruct-q4f16_1-MLC",
model_lib="./dist/prebuilt/lib/Llama-3-8b-Instruct/Llama-3-8B-Instruct-q4f16_1-mali.so",
device="opencl"
engine = MLCEngine(model=model, model_lib=model_lib, device=device)

# Run chat completion in OpenAI API.
for response in engine.chat.completions.create(
    messages=[{"role": "user", "content": "What is the meaning of life?"}],
    stream=True,
):
    for choice in response.choices:
        print(choice.delta.content, end="", flush=True)
print("\n")

engine.terminate()

Output:

Traceback (most recent call last):
  File "/home/hank/repos/mlc-llm/test.py", line 7, in <module>
    engine = MLCEngine(model=model, model_lib=model_lib, device=device)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/hank/repos/mlc-llm/python/mlc_llm/serve/engine.py", line 1466, in __init__
    super().__init__(
  File "/home/hank/repos/mlc-llm/python/mlc_llm/serve/engine_base.py", line 590, in __init__
    ) = _process_model_args(models, device, engine_config)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/hank/repos/mlc-llm/python/mlc_llm/serve/engine_base.py", line 171, in _process_model_args
    model_args: List[Tuple[str, str]] = [_convert_model_info(model) for model in models]
                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/hank/repos/mlc-llm/python/mlc_llm/serve/engine_base.py", line 171, in <listcomp>
    model_args: List[Tuple[str, str]] = [_convert_model_info(model) for model in models]
                                         ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/hank/repos/mlc-llm/python/mlc_llm/serve/engine_base.py", line 124, in _convert_model_info
    model_path = download_cache.get_or_download_model(model.model)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/hank/repos/mlc-llm/python/mlc_llm/support/download_cache.py", line 226, in get_or_download_model
    if model.startswith("HF://"):
       ^^^^^^^^^^^^^^^^
AttributeError: 'tuple' object has no attribute 'startswith'

PDB run:

> /home/hank/repos/mlc-llm/python/mlc_llm/support/download_cache.py(227)get_or_download_model()
-> if model.startswith("HF://"):
(Pdb) model
('./dist/prebuilt/Llama-3-8B-Instruct-q4f16_1-MLC',)

Expected behavior

Expected the local model to be loaded correctly as this use of MLCEngine directly matches the documentation.

Environment

• Platform (e.g. WebGPU/Vulkan/IOS/Android/CUDA): OpenCL
• Operating system (e.g. Ubuntu/Windows/MacOS/...): Armbian
• Device (e.g. iPhone 12 Pro, PC+RTX 3090, ...): OrangePi 5 Max 16GB
• How you installed MLC-LLM (conda, source): source
• How you installed TVM-Unity (pip, source): source
• Python version (e.g. 3.10): 3.11.11
• GPU driver version (if applicable): N/A
• CUDA/cuDNN version (if applicable): N/A
• TVM Unity Hash Tag: d6c1489b5d030c03fa9132accdb2d72f237f3e3e
• Any other relevant information: N/A

Thanks! The model does work when using HF:// URLs.