arjunguha commited on
Commit
6d6272d
1 Parent(s): 7b6337d

Version 2: added MBPP

Browse files
Files changed (3) hide show
  1. MultiPL-E.py +24 -6
  2. README.md +5 -3
  3. dataset_infos.json +0 -0
MultiPL-E.py CHANGED
@@ -23,10 +23,13 @@ _CITATION = """\
23
  _DESCRIPTION = """\
24
  MultiPL-E is a dataset for evaluating large language models for code \
25
  generation that supports 18 programming languages. It takes the OpenAI \
26
- "HumanEval" Python benchmarks and uses little compilers to translate them \
27
- to other languages. It is easy to add support for new languages and benchmarks.
 
28
  """
29
 
 
 
30
  _LANGUAGES = [
31
  "cpp", "cs", "d", "go", "java", "jl", "js", "lua", "php", "pl", "py", "r",
32
  "rb", "rkt", "rs", "scala", "sh", "swift", "ts"
@@ -39,29 +42,44 @@ class MultiPLEBuilderConfig(datasets.BuilderConfig):
39
 
40
  def __init__(
41
  self,
 
42
  language,
43
  variation,
44
  **kwargs,
45
  ):
46
  self.language = language
47
  self.variation = variation
48
- name = f"{language}-{variation}" if variation != "reworded" else language
 
 
 
49
  kwargs["name"] = name
50
  super(MultiPLEBuilderConfig, self).__init__(**kwargs)
51
 
 
 
 
 
 
 
 
 
52
  class MultiPLE(datasets.GeneratorBasedBuilder):
53
  BUILDER_CONFIG_CLASS = MultiPLEBuilderConfig
54
 
55
  BUILDER_CONFIGS = [
56
  MultiPLEBuilderConfig(
 
57
  language=language,
58
  variation=variation,
59
- version=datasets.Version("1.0.0"))
 
60
  for language in _LANGUAGES
61
  for variation in _VARIATIONS
 
62
  ]
63
 
64
- DEFAULT_CONFIG_NAME = "cpp-reworded"
65
 
66
  def _info(self):
67
  return datasets.DatasetInfo(
@@ -85,7 +103,7 @@ class MultiPLE(datasets.GeneratorBasedBuilder):
85
 
86
  def _split_generators(self, dl_manager: datasets.DownloadManager):
87
  files = dl_manager.download(
88
- f"https://raw.githubusercontent.com/nuprl/MultiPL-E/375e903198713b7f5faa95a4047c6928cf7348f9/prompts/{self.config.language}-{self.config.variation}.json"
89
  )
90
  return [
91
  datasets.SplitGenerator(
 
23
  _DESCRIPTION = """\
24
  MultiPL-E is a dataset for evaluating large language models for code \
25
  generation that supports 18 programming languages. It takes the OpenAI \
26
+ "HumanEval" and the MBPP Python benchmarks and uses little compilers to \
27
+ translate them to other languages. It is easy to add support for new languages \
28
+ and benchmarks.
29
  """
30
 
31
+ _SRCDATA = [ "humaneval", "mbpp" ]
32
+
33
  _LANGUAGES = [
34
  "cpp", "cs", "d", "go", "java", "jl", "js", "lua", "php", "pl", "py", "r",
35
  "rb", "rkt", "rs", "scala", "sh", "swift", "ts"
 
42
 
43
  def __init__(
44
  self,
45
+ srcdata,
46
  language,
47
  variation,
48
  **kwargs,
49
  ):
50
  self.language = language
51
  self.variation = variation
52
+ self.srcdata = srcdata
53
+ name = f"{srcdata}-{language}"
54
+ if variation != "reworded":
55
+ name = f"{name}-{variation}"
56
  kwargs["name"] = name
57
  super(MultiPLEBuilderConfig, self).__init__(**kwargs)
58
 
59
+ def _is_interesting(srcdata: str, variation: str):
60
+ if srcdata == "humaneval":
61
+ return True
62
+ if srcdata == "mbpp":
63
+ # MBPP does not have doctests, so these are the only interesting
64
+ # variations
65
+ return variation in [ "keep", "reworded" ]
66
+
67
  class MultiPLE(datasets.GeneratorBasedBuilder):
68
  BUILDER_CONFIG_CLASS = MultiPLEBuilderConfig
69
 
70
  BUILDER_CONFIGS = [
71
  MultiPLEBuilderConfig(
72
+ srcdata=srcdata,
73
  language=language,
74
  variation=variation,
75
+ version=datasets.Version("2.0.0"))
76
+ for srcdata in _SRCDATA
77
  for language in _LANGUAGES
78
  for variation in _VARIATIONS
79
+ if _is_interesting(srcdata, variation)
80
  ]
81
 
82
+ DEFAULT_CONFIG_NAME = "humaneval-cpp"
83
 
84
  def _info(self):
85
  return datasets.DatasetInfo(
 
103
 
104
  def _split_generators(self, dl_manager: datasets.DownloadManager):
105
  files = dl_manager.download(
106
+ f"https://raw.githubusercontent.com/nuprl/MultiPL-E/1f21818a0f3265fd0a41c3954e30aab47f34063a/prompts/{self.config.srcdata}-{self.config.language}-{self.config.variation}.json"
107
  )
108
  return [
109
  datasets.SplitGenerator(
README.md CHANGED
@@ -16,6 +16,7 @@ size_categories:
16
  source_datasets:
17
  - original
18
  - extended|openai_humaneval
 
19
  tags: []
20
  task_categories: []
21
  task_ids: []
@@ -34,8 +35,9 @@ task_ids: []
34
 
35
  MultiPL-E is a dataset for evaluating large language models for code
36
  generation that supports 18 programming languages. It takes the OpenAI
37
- "HumanEval" Python benchmarks and uses little compilers to translate them
38
- to other languages. It is easy to add support for new languages and benchmarks.
 
39
 
40
  ## Example
41
 
@@ -50,7 +52,7 @@ LANG = "lua"
50
  MODEL_NAME = "Salesforce/codegen-350M-multi"
51
  tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
52
  model = AutoModelForCausalLM.from_pretrained(MODEL_NAME).half().cuda()
53
- problems = datasets.load_dataset("nuprl/MultiPL-E", LANG)
54
 
55
  def stop_at_stop_token(decoded_string, problem):
56
  """
 
16
  source_datasets:
17
  - original
18
  - extended|openai_humaneval
19
+ - extended|mbpp
20
  tags: []
21
  task_categories: []
22
  task_ids: []
 
35
 
36
  MultiPL-E is a dataset for evaluating large language models for code
37
  generation that supports 18 programming languages. It takes the OpenAI
38
+ "HumanEval" and the MBPP Python benchmarks and uses little compilers to
39
+ translate them to other languages. It is easy to add support for new languages
40
+ and benchmarks.
41
 
42
  ## Example
43
 
 
52
  MODEL_NAME = "Salesforce/codegen-350M-multi"
53
  tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
54
  model = AutoModelForCausalLM.from_pretrained(MODEL_NAME).half().cuda()
55
+ problems = datasets.load_dataset("nuprl/MultiPL-E", f"humaneval-{LANG}")
56
 
57
  def stop_at_stop_token(decoded_string, problem):
58
  """
dataset_infos.json CHANGED
The diff for this file is too large to render. See raw diff