| import json |
| from pathlib import Path |
| import pytest |
| from transformers import AutoTokenizer |
| from v4_prompt import messages_for |
| from core import normalize_question,text |
| from openjudgement import OpenJudgement |
| ROOT=Path(__file__).resolve().parents[1] |
|
|
| def test_real_tokenizer_and_examples(): |
| tokenizer=AutoTokenizer.from_pretrained(ROOT,local_files_only=True) |
| assert all(len(tokenizer.encode(str(i),add_special_tokens=False))==1 for i in range(10)) |
| assert len(tokenizer.encode('10',add_special_tokens=False))>1 |
| req=json.loads((ROOT/'examples/request.json').read_text()) |
| for q in req['questions'].values(): |
| kind,instruction,options,keys=normalize_question(q) |
| row={'state':text(req['state']),'kind':kind,'instructions':instruction,'candidates':options,'keys':keys} |
| tokens=tokenizer.apply_chat_template(messages_for(row),tokenize=True,add_generation_prompt=True,enable_thinking=False,return_dict=False) |
| assert isinstance(tokens,list) and 0<len(tokens)<32768 |
|
|
| def test_loader_local_metadata_without_weights(tmp_path,monkeypatch): |
| (tmp_path/'config.json').write_text('{}') |
| (tmp_path/'openjudgment_config.json').write_text(json.dumps({'model_id':'kitaniai/OpenJudgement-4B-Preview'})) |
| called={} |
| def fake_init(self,path,**kwargs): called.update(path=path,**kwargs) |
| monkeypatch.setattr(OpenJudgement,'__init__',fake_init) |
| monkeypatch.setattr(OpenJudgement,'eval',lambda self:self) |
| judge=OpenJudgement.from_pretrained(str(tmp_path),device='cpu') |
| assert judge.model_name=='kitaniai/OpenJudgement-4B-Preview' |
| assert called['device']=='cpu' |
|
|
| def test_loader_rejects_incomplete_local_folder(tmp_path): |
| with pytest.raises(ValueError,match='config.json'): |
| OpenJudgement.from_pretrained(str(tmp_path),device='cpu') |
|
|
