Do you support running in an mps environment?

[zhangzhenxi]

I see that readme only provides guidelines for running under cuda and cpu. Do you support mac mps devices?

[darkrush]

Our project is primarily developed on Linux systems with CUDA GPU support. We currently lack a cross-platform development environment, so there are no immediate plans for cross-platform development.

As an alternative solution, we will modify the API call interface to allow users to pass their own locally executable instance of Qwen3-0.6B inference as the backend for this tool. This will partially alleviate cross-platform compatibility issues.

[zhangzhenxi]

Great, look forward to your updates

[darkrush]

We will also soon launch a free API on MinerU.net, which includes the latest content extraction model + enhanced HTML-to-Markdown functionality. It is expected to go live around January 2026. Stay tuned!

[ideaflow]

We have recently added an abstract inference backend class in dripper/inference/imp.py. You can inherit from it to implement a concrete inference backend class.

We have already implemented inference backend classes for VLLM and Hugging Face Transformers. You can refer to these examples to implement an inference backend class for Mac MPS, for instance:

class MPSInferenceBackend(InferenceBackend):
    def __init__(self, model_path: str, model_init_kwargs: Dict[str, Any] = {}, model_gen_kwargs: Dict[str, Any] = {}):
        # Concrete instantiation code here

    @override
    def generate(self, prompt_list: list[str]) -> list[ModelResponse]:
        model_output = ...  # Concrete inference logic here
        response = []
        for res in model_output:
            response.append(
                ModelResponse(
                    generated_text=...
                )
            )
        return response

Then, in dripper/api.py, locate the get_llm method within the Dripper class and add a conditional branch to instantiate the MPSInferenceBackend class, for example:

from dripper.inference.imp import MPSInferenceBackend
...
class Dripper:
    ...
    def get_llm(self) -> InferenceBackend:
        ...
        # Your new code >>
        elif self.inference_backend == 'mps':
            self._llm = MPSInferenceBackend(
                model_path=self.model_path,
                model_init_kwargs=self.model_init_kwargs,
                model_gen_kwargs=self.model_gen_kwargs
            )
        # << Your new code
        else:
            raise DripperConfigError(
                f'Unsupported inference backend: {self.inference_backend}'
            )
        ...

Finally, refer to the latest README to instantiate the Dripper class:

dripper = Dripper(
    config={
        'model_path': '/path/to/your/model',
        'use_fall_back': True,
        'raise_errors': False,
        'inference_backend': 'mps',
        'model_init_kwargs': {
            ...
        },
        'model_gen_kwargs': {
            ...
        },
    }
)
html_content = ...
result = dripper.process(html_content)

Hopefully this will meet your needs.

[zhangzhenxi]

That's great! I'm really excited to see where this goes. If I need any support along the way, I’ll be sure to reach out and keep you posted.