/
cr_wwo_reprompt.py
55 lines (45 loc) · 2.36 KB
/
cr_wwo_reprompt.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
import json
import pathlib
import itertools
from anthropic import BadRequestError
from src.datasets import load_dataset
from src.models import load_model
from src.run import run
def saferun(*args, **kwargs):
try:
return run(*args, **kwargs)
except BadRequestError as e:
if 'Output blocked by content filtering policy' in e.message:
out = {'response': 'CONTENT BLOCKED'}
pathlib.Path(kwargs['output_path']).parent.mkdir(parents=True, exist_ok=True)
with open(kwargs['output_path'], 'w') as f:
json.dump(out, f, indent=2)
return None
except:
return None
model_name = 'gpt4'
if model_name == 'gpt4':
model = load_model('openai')
if model_name == 'claude':
model = load_model('anthropic')
dataset_names = ['nq', 'squad', 'hotpotqa', 'pubmed']
context_len = 80000
answer_positions = range(0, context_len+1, 10000)
for dataset_name in dataset_names:
dataset = load_dataset(dataset_name)
question_ids = range(250) if dataset_name == 'hotpotqa' else range(50)
for (question_id, answer_position) in itertools.product(question_ids, answer_positions):
if answer_position > 0 and dataset_name == 'hotpotqa':
continue
print(f'{dataset_name} {context_len//1000:d}k q{question_id} a{answer_position}')
for cs in [10, 20, 40, 80]:
print(f'CR {cs:d}k')
out = saferun(model, dataset, question_id=question_id, answer_position=answer_position, total_context=context_len, return_page=True, do_abbreviate=True, do_chunking=True, chunk_size=cs*1000, output_path=f'results/cvr/{dataset_name}/{model_name}/c{context_len:d}/cr{cs:d}k/q{question_id:d}/a{answer_position:d}.json')
if cs == 10:
continue
print(f'CR {cs:d}k + Reprompt')
if cs == 80:
out = saferun(model, dataset, question_id=question_id, answer_position=answer_position, total_context=context_len, return_page=True, repeat_prompt=True, repeat_interval=10000, do_abbreviate=True, output_path=f'results/cvr/{dataset_name}/{model_name}/c{context_len:d}/cr80k+reprompt/q{question_id:d}/a{answer_position:d}.json')
else:
out = saferun(model, dataset, question_id=question_id, answer_position=answer_position, total_context=context_len, return_page=True, repeat_prompt=True, repeat_interval=10000, do_abbreviate=True, do_chunking=True, chunk_size=cs*1000, output_path=f'results/cvr/{dataset_name}/{model_name}/c{context_len:d}/cr{cs:d}k+reprompt/q{question_id:d}/a{answer_position:d}.json')
print('done!')