lambada.py 2.02 KB
Newer Older
1
import json
&'s avatar
& committed
2
3
from lm_eval.base import Task, rf
from lm_eval.metrics import mean, perplexity
sdtblck's avatar
sdtblck committed
4
from lm_eval.utils import sh
Leo Gao's avatar
Leo Gao committed
5
from best_download import download_file
sdtblck's avatar
sdtblck committed
6
7


8
class LAMBADA(Task):
Leo Gao's avatar
Leo Gao committed
9
    VERSION = 0
sdtblck's avatar
sdtblck committed
10
11
    def download(self):
        sh("mkdir -p data/lambada")
12
13
14
15
16
17
18
19
20
21
        try:
            download_file(
                "http://eaidata.bmk.sh/data/lambada_test.jsonl", 
                "data/lambada/lambada_test.jsonl", 
                "4aa8d02cd17c719165fc8a7887fddd641f43fcafa4b1c806ca8abc31fabdb226"
            )
        except:
            # fallback - for some reason best_download doesnt work all the time here
            sh("wget http://eaidata.bmk.sh/data/lambada_test.jsonl -O data/lambada/lambada_test.jsonl")
            sh('echo "4aa8d02cd17c719165fc8a7887fddd641f43fcafa4b1c806ca8abc31fabdb226  data/lambada/lambada_test.jsonl" | sha256sum --check')
sdtblck's avatar
sdtblck committed
22
23
24
25
26

    def has_training_docs(self):
        return False

    def has_validation_docs(self):
Leo Gao's avatar
Leo Gao committed
27
        return True
sdtblck's avatar
sdtblck committed
28
29

    def has_test_docs(self):
Leo Gao's avatar
Leo Gao committed
30
        return False
sdtblck's avatar
sdtblck committed
31
32
33
34
35

    def training_docs(self):
        pass

    def validation_docs(self):
Leo Gao's avatar
Leo Gao committed
36
37
38
39
        with open("data/lambada/lambada_test.jsonl") as fh:
            for line in fh:
                yield json.loads(line)

Leo Gao's avatar
Leo Gao committed
40
41
42
    def test_docs(self):
        pass

Leo Gao's avatar
Leo Gao committed
43
44
    def doc_to_text(self, doc):
        return doc['text'].rsplit(' ', 1)[0]
sdtblck's avatar
sdtblck committed
45

Leo Gao's avatar
Leo Gao committed
46
47
    def doc_to_target(self, doc):
        return " " + doc['text'].rsplit(' ', 1)[1]
Leo Gao's avatar
Leo Gao committed
48
49
50
51
    
    def fewshot_description(self):
        # TODO: figure out description
        return ""
sdtblck's avatar
sdtblck committed
52

Leo Gao's avatar
Leo Gao committed
53
    def construct_requests(self, doc, ctx):
Leo Gao's avatar
Leo Gao committed
54
        ll, is_greedy = rf.loglikelihood(ctx, self.doc_to_target(doc))
55

Leo Gao's avatar
Leo Gao committed
56
        return ll, is_greedy
Leo Gao's avatar
Leo Gao committed
57
58
    
    def process_results(self, doc, results):
Leo Gao's avatar
Leo Gao committed
59
        ll, is_greedy = results
Leo Gao's avatar
Leo Gao committed
60

Leo Gao's avatar
Leo Gao committed
61
        return {
Leo Gao's avatar
Leo Gao committed
62
63
            'ppl': ll,
            'acc': int(is_greedy)
Leo Gao's avatar
Leo Gao committed
64
65
        }
        
Leo Gao's avatar
Leo Gao committed
66
    def aggregation(self):
Leo Gao's avatar
Leo Gao committed
67
        return {
Leo Gao's avatar
Leo Gao committed
68
69
            'ppl': perplexity,
            'acc': mean
Leo Gao's avatar
Leo Gao committed
70
        }
Leo Gao's avatar
Leo Gao committed
71
72

    def higher_is_better(self):
Leo Gao's avatar
Leo Gao committed
73
        return {
Leo Gao's avatar
Leo Gao committed
74
75
            'ppl': False,
            'acc': True
Leo Gao's avatar
Leo Gao committed
76
        }