Version 2: added MBPP

MultiPL-E.py

@@ -23,10 +23,13 @@ _CITATION = """\
 _DESCRIPTION = """\
 MultiPL-E is a dataset for evaluating large language models for code \
 generation that supports 18 programming languages. It takes the OpenAI \
-"HumanEval" Python benchmarks and uses little compilers to translate them \
-to other languages. It is easy to add support for new languages and benchmarks.
+"HumanEval" and the MBPP Python benchmarks and uses little compilers to \
+translate them to other languages. It is easy to add support for new languages \
+and benchmarks.
 """
 
+_SRCDATA = [ "humaneval", "mbpp" ]
+
 _LANGUAGES = [ 
     "cpp", "cs", "d", "go", "java", "jl", "js", "lua", "php", "pl", "py", "r", 
     "rb", "rkt", "rs", "scala", "sh", "swift", "ts"
@@ -39,29 +42,44 @@ class MultiPLEBuilderConfig(datasets.BuilderConfig):
 
     def __init__(
         self,
+        srcdata,
         language,
         variation,
         **kwargs,
     ):
         self.language = language
         self.variation = variation
-        name = f"{language}-{variation}" if variation != "reworded" else language
+        self.srcdata = srcdata
+        name = f"{srcdata}-{language}"
+        if variation != "reworded":
+            name = f"{name}-{variation}"
         kwargs["name"] = name
         super(MultiPLEBuilderConfig, self).__init__(**kwargs)
 
+def _is_interesting(srcdata: str, variation: str):
+    if srcdata == "humaneval":
+        return True
+    if srcdata == "mbpp":
+        # MBPP does not have doctests, so these are the only interesting
+        # variations
+        return variation in [ "keep", "reworded" ]
+
 class MultiPLE(datasets.GeneratorBasedBuilder):
     BUILDER_CONFIG_CLASS = MultiPLEBuilderConfig
 
     BUILDER_CONFIGS = [
         MultiPLEBuilderConfig(
+            srcdata=srcdata,
             language=language,
             variation=variation,
-            version=datasets.Version("1.0.0"))
+            version=datasets.Version("2.0.0"))
+        for srcdata in _SRCDATA
         for language in _LANGUAGES 
         for variation in _VARIATIONS
+        if _is_interesting(srcdata, variation)
     ]
 
-    DEFAULT_CONFIG_NAME = "cpp-reworded"
+    DEFAULT_CONFIG_NAME = "humaneval-cpp"
 
     def _info(self):
         return datasets.DatasetInfo(
@@ -85,7 +103,7 @@ class MultiPLE(datasets.GeneratorBasedBuilder):
 
     def _split_generators(self, dl_manager: datasets.DownloadManager):
         files = dl_manager.download(
-            f"https://raw.githubusercontent.com/nuprl/MultiPL-E/375e903198713b7f5faa95a4047c6928cf7348f9/prompts/{self.config.language}-{self.config.variation}.json"
+            f"https://raw.githubusercontent.com/nuprl/MultiPL-E/1f21818a0f3265fd0a41c3954e30aab47f34063a/prompts/{self.config.srcdata}-{self.config.language}-{self.config.variation}.json"
         )
         return [
             datasets.SplitGenerator(

README.md

@@ -16,6 +16,7 @@ size_categories:
 source_datasets:
 - original
 - extended|openai_humaneval
+- extended|mbpp
 tags: []
 task_categories: []
 task_ids: []
@@ -34,8 +35,9 @@ task_ids: []
 
 MultiPL-E is a dataset for evaluating large language models for code
 generation that supports 18 programming languages. It takes the OpenAI 
-"HumanEval" Python benchmarks and uses little compilers to translate them 
-to other languages. It is easy to add support for new languages and benchmarks.
+"HumanEval" and the MBPP Python benchmarks and uses little compilers to
+translate them  to other languages. It is easy to add support for new languages 
+and benchmarks.
 
 ## Example
 
@@ -50,7 +52,7 @@ LANG = "lua"
 MODEL_NAME = "Salesforce/codegen-350M-multi"
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 model = AutoModelForCausalLM.from_pretrained(MODEL_NAME).half().cuda()
-problems = datasets.load_dataset("nuprl/MultiPL-E", LANG)
+problems = datasets.load_dataset("nuprl/MultiPL-E", f"humaneval-{LANG}")
 
 def stop_at_stop_token(decoded_string, problem):
     """