Function do_generate

utils/src/experiment/experiments.py:88–128

(
    genclass: Type[PPTCrew],
    setting: str,
    model_identifier: str,
    debug: bool,
    ppt_folder: str,
    thread_id: int,
    num_slides: int = 12,
)

def do_generate(
    genclass: Type[PPTCrew],
    setting: str,
    model_identifier: str,
    debug: bool,
    ppt_folder: str,
    thread_id: int,
    num_slides: int = 12,
):
    app_config = Config(rundir=ppt_folder, debug=debug)
    text_model = get_text_model(f"cuda:{thread_id % torch.cuda.device_count()}")
    presentation = Presentation.from_file(
        pjoin(ppt_folder, "source.pptx"),
        app_config,
    )
    ImageLabler(presentation, app_config).caption_images()
    induct_cache = pjoin(
        app_config.RUN_DIR, "template_induct", model_identifier, "induct_cache.json"
    )
    if not older_than(induct_cache, wait=True):
        print(f"induct_cache not found: {induct_cache}")
        return
    slide_induction = json.load(open(induct_cache))
    try:
        pptgen: PPTCrew = genclass(text_model).set_reference(presentation, slide_induction)
    except:
        print("set_reference failed")
        pptgen: PPTCrew = genclass(text_model).set_reference(presentation, slide_induction)
        
    topic = ppt_folder.split("/")[1]
    for pdf_folder in glob(f"data/{topic}/pdf/*"):
        app_config.set_rundir(pjoin(ppt_folder, setting, pbasename(pdf_folder)))
        if pexists(pjoin(app_config.RUN_DIR, "history")):
            continue
        images = json.load(
            open(pjoin(pdf_folder, "image_caption.json"), "r"),
        )
        doc_json = json.load(
            open(pjoin(pdf_folder, "refined_doc.json"), "r"),
        )
        pptgen.generate_pres(app_config, images, num_slides, doc_json)


def generate_pres(