(induct_id: int, wait: bool = False)
| 203 | |
| 204 | |
| 205 | def prepare_induction(induct_id: int, wait: bool = False): |
| 206 | induct_llms = [ |
| 207 | (llms.qwen2_5, llms.qwen_vl), |
| 208 | (llms.gpt4o, llms.gpt4o), |
| 209 | (llms.qwen_vl, llms.qwen_vl), |
| 210 | ] |
| 211 | |
| 212 | def do_induct(llm: list[llms.LLM], ppt_folder: str, rank: int): |
| 213 | if not older_than(pjoin(ppt_folder, "source.pptx"), wait=wait): |
| 214 | return |
| 215 | llms.language_model = llm[0] |
| 216 | llms.vision_model = llm[1] |
| 217 | config = Config(rundir=ppt_folder) |
| 218 | ppt_image_folder = pjoin(ppt_folder, "source_slides") |
| 219 | template_image_folder = pjoin(ppt_folder, "template_images") |
| 220 | image_model = get_image_model(f"cuda:{rank % device_count}") |
| 221 | presentation = Presentation.from_file(pjoin(ppt_folder, "source.pptx"), config) |
| 222 | ImageLabler(presentation, config).caption_images() |
| 223 | slide_inducter = SlideInducter( |
| 224 | presentation, ppt_image_folder, template_image_folder, config, image_model |
| 225 | ) |
| 226 | slide_inducter.content_induct() |
| 227 | |
| 228 | for folder in tqdm(sorted(glob.glob("data/*/pptx/*")), desc="prepare induction"): |
| 229 | do_induct(induct_llms[induct_id], folder, 0) |
| 230 | |
| 231 | |
| 232 | if __name__ == "__main__": |
no test coverage detected