| 37 | |
| 38 | # === General VLM Inference Interface === |
| 39 | class VLM(Protocol): |
| 40 | image_processor: ImageProcessor |
| 41 | |
| 42 | def load(self) -> Tuple[nn.Module, Tokenizer, ImageProcessor]: |
| 43 | ... |
| 44 | |
| 45 | def get_prompt_builder(self, system_prompt: Optional[str] = None) -> Any: |
| 46 | ... |
| 47 | |
| 48 | def get_prompt_fn(self, dataset_family: str = "vqa-v2") -> Callable[[str], str]: |
| 49 | ... |
| 50 | |
| 51 | def generate_answer( |
| 52 | self, |
| 53 | pixel_values: torch.Tensor, |
| 54 | question_prompts: List[str], |
| 55 | return_string_probabilities: Optional[List[str]] = None, |
| 56 | ) -> Union[List[str], List[List[float]]]: |
| 57 | ... |
| 58 | |
| 59 | def generate( |
| 60 | self, |
| 61 | image: Image, |
| 62 | input_text: str, |
| 63 | do_sample: bool, |
| 64 | temperature: float, |
| 65 | max_new_tokens: int, |
| 66 | min_length: int, |
| 67 | length_penalty: float, |
| 68 | **kwargs, |
| 69 | ) -> str: |
| 70 | ... |
nothing calls this directly
no outgoing calls
no test coverage detected