MCPcopy Create free account
hub / github.com/CoopReason/TESSY / async_main

Function async_main

run_tessy.py:361–645  ·  view source on GitHub ↗
(args)

Source from the content-addressed store, hash-verified

359
360
361async def async_main(args):
362 # Classifier model loading still in main, then passed in
363 teacher_api_model_name = args.api_model_name_teacher
364 student_api_model_name = args.api_model_name_student
365 teacher_api_url = args.teacher_api_url
366 student_api_url = args.student_api_url
367
368 input_path = args.input_path
369 output_path = args.output_path
370 enable_think = args.enable_thinking
371 classifier_len = args.classifier_len # Use local variable
372 max_new_tokens_per_sample = args.max_new_tokens
373
374
375 system_prompt = args.system_prompt
376
377 dataset = read_jsonl(input_path)
378 processed_ids = load_processed_ids(output_path)
379
380 llm_tokenizer_teacher = AutoTokenizer.from_pretrained(args.model_name_teacher, trust_remote_code=True)
381 if llm_tokenizer_teacher.pad_token is None:
382 llm_tokenizer_teacher.pad_token = llm_tokenizer_teacher.eos_token
383
384 llm_tokenizer_student = AutoTokenizer.from_pretrained(args.model_name_student, trust_remote_code=True)
385 if llm_tokenizer_student.pad_token is None:
386 llm_tokenizer_student.pad_token = llm_tokenizer_student.eos_token
387
388 # Load teacher classifier tokenizer and model
389 teacher_token_classifier_tokenizer = AutoTokenizer.from_pretrained(args.teacher_classifier_path, trust_remote_code=True)
390 teacher_token_classifier_tokenizer.truncation_side = "left"
391 teacher_classifier_config = AutoConfig.from_pretrained(args.teacher_classifier_path, num_labels=2)
392 teacher_classifier_model_base = AutoModelForTokenClassification.from_pretrained(
393 args.teacher_classifier_path, config=teacher_classifier_config, ignore_mismatched_sizes=True, trust_remote_code=True
394 )
395
396 print('====================')
397 print('Loading teacher classifier from:', args.teacher_classifier_path)
398 print('====================')
399
400 teacher_classifier_model_base = teacher_classifier_model_base.bfloat16().cuda()
401 teacher_token_classifier_model = nn.DataParallel(teacher_classifier_model_base).eval()
402
403 # Load student classifier tokenizer and model
404 student_token_classifier_tokenizer = AutoTokenizer.from_pretrained(args.student_classifier_path, trust_remote_code=True)
405 student_token_classifier_tokenizer.truncation_side = "left"
406 student_classifier_config = AutoConfig.from_pretrained(args.student_classifier_path, num_labels=2)
407 student_classifier_model_base = AutoModelForTokenClassification.from_pretrained(
408 args.student_classifier_path, config=student_classifier_config, ignore_mismatched_sizes=True, trust_remote_code=True
409 )
410 student_classifier_model_base = student_classifier_model_base.bfloat16().cuda()
411 print('====================')
412 print('Loading student classifier from:', args.student_classifier_path)
413 print('====================')
414 student_token_classifier_model = nn.DataParallel(student_classifier_model_base).eval()
415
416
417
418 dataset_queue = deque(dataset) # Use deque to support queue tail append

Callers 1

run_tessy.pyFile · 0.85

Calls 8

post_process_textFunction · 0.85
read_jsonlFunction · 0.70
load_processed_idsFunction · 0.70
get_hashes_and_linesFunction · 0.70
build_promptFunction · 0.70
append_jsonlFunction · 0.70

Tested by

no test coverage detected