(model_out)
| 360 | bp = 0 |
| 361 | b_div = 0 |
| 362 | def work(model_out): |
| 363 | global ask, tell, lev_info, auv, lock, logger, entities |
| 364 | global total_informational_div, informational, support_info |
| 365 | global br, bf, bp, b_div |
| 366 | if isinstance(model_out['model'], list): |
| 367 | model_out['model'] = model_out['model'][0] |
| 368 | r, p, f = score([ |
| 369 | model_out["model"], |
| 370 | ], [model_out['src']['conversation'][-1]['doctor']], lang='zh') |
| 371 | br += r[0] |
| 372 | bf += p[0] |
| 373 | bp += f[0] |
| 374 | b_div += 1 |
| 375 | logger.info(model_out) |
| 376 | if is_ask(model_out["model"]): |
| 377 | model_entities = generate_entities(model_out["model"]) |
| 378 | actual_resp = model_out['src']['conversation'][-1]['doctor'] |
| 379 | actual_entities = generate_entities(actual_resp) |
| 380 | t_auv = get_auv(model_out, model_entities) |
| 381 | lev_d = Levenshtein.ratio( |
| 382 | flatten_medical_ask_content(actual_entities), |
| 383 | flatten_medical_ask_content(model_entities), |
| 384 | ) |
| 385 | entities.append( |
| 386 | (actual_entities, model_entities) |
| 387 | ) |
| 388 | if True: |
| 389 | ask += 1 |
| 390 | lev_info.append(lev_d) |
| 391 | auv.append( |
| 392 | t_auv |
| 393 | ) |
| 394 | else: |
| 395 | tell += 1 |
| 396 | actual_resp = model_out['src']['conversation'][-1]['doctor'] |
| 397 | is_f = False |
| 398 | if get_is_informational( |
| 399 | model_out['src']['conversation'][-1]['patient'], |
| 400 | model_out["model"], |
| 401 | actual_resp, |
| 402 | model_out['src']['conversation'][:-1]): |
| 403 | informational += 1 |
| 404 | is_f = True |
| 405 | total_informational_div += 1 |
| 406 | if is_f: |
| 407 | if not hallucinate( |
| 408 | model_out['src']['conversation'][-1]['patient'], |
| 409 | model_out["model"], |
| 410 | actual_resp, |
| 411 | model_out['src']['conversation'][:-1]): |
| 412 | support_info.append(1) |
| 413 | else: |
| 414 | support_info.append(0) |
| 415 | futures = [] |
| 416 | for i in tqdm(data[:]): |
| 417 | work(i) |
no test coverage detected