| import transformers |
| from transformers.models.auto import CONFIG_MAPPING |
|
|
|
|
| class CXRMateEDConfig(transformers.PretrainedConfig): |
|
|
| model_type = 'cxrmate-ed' |
|
|
| def __init__( |
| self, |
| vision_config=None, |
| text_config=None, |
| index_value_encoder_intermediate_size: int = 2048, |
| include_time_delta: bool = True, |
| time_delta_monotonic_inversion: bool = True, |
| add_time_deltas: bool = True, |
| history: int = 0, |
| tables_filter: list = ['mimic_cxr_sectioned', 'triage', 'medrecon'], |
| prompt_report_sections_filter: list = ['indication', 'history'], |
| pad_token_id: int = 4, |
| **kwargs, |
| ): |
| super().__init__(**kwargs) |
| self.vision_config = vision_config |
| |
| self.index_value_encoder_intermediate_size = index_value_encoder_intermediate_size |
| self.include_time_delta = include_time_delta |
| self.time_delta_monotonic_inversion = time_delta_monotonic_inversion |
| self.add_time_deltas = add_time_deltas |
| self.history = history |
| self.tables_filter = tables_filter |
| self.prompt_report_sections_filter = prompt_report_sections_filter |
| self.pad_token_id = pad_token_id |
|
|
| if isinstance(vision_config, dict): |
| vision_config = transformers.AutoConfig.from_pretrained( |
| 'aehrc/uniformer_base_tl_384', |
| trust_remote_code=True, |
| **vision_config, |
| ) |
| |
| self.vision_config = vision_config |
|
|
| if isinstance(text_config, dict): |
| text_config['model_type'] = text_config['model_type'] if 'model_type' in text_config else 'llama' |
| text_config = CONFIG_MAPPING[text_config['model_type']](**text_config) |
|
|
| self.text_config = text_config |
|
|
