implemented flores200

lm_eval/tasks/opengptx/all_tasks_registry.py

@@ -1,4 +1,5 @@
 # OpenGPT-X tasks
+from . import flores200
 from . import german_europarl_ppl
 from . import german_ler_ppl
 from . import germanquad
@@ -17,6 +18,12 @@
 from . import xquad
 from . import xnli
 
+########################################
+# Translation tasks
+########################################
+
+euro5_flores_benchmark = ["eng_Latn", "deu_Latn", "fra_Latn", "ita_Latn", "spa_Latn"]
+
 
 TASK_REGISTRY_TMP = {
     # OpenGPT-X tasks
@@ -40,6 +47,7 @@
     "xstance_fr": x_stance.XStanceFR,
     **xquad.construct_tasks(),
     **xnli.construct_tasks(),
+    **flores200.create_tasks_from_list(euro5_flores_benchmark)
 }
 
 # add a prefix to tasks implemented by OpenGPT-X

lm_eval/tasks/opengptx/flores200.py

@@ -0,0 +1,120 @@
+"""
+NOTE: This file implements the Flores200 translation task, see
+https://github.com/facebookresearch/flores/tree/main/flores200.
+"""
+import pycountry
+import datasets
+from itertools import permutations
+from pprint import pprint
+from sacrebleu import sacrebleu
+from lm_eval import metrics
+from lm_eval.base import Task, rf
+from lm_eval.tasks.translation import code_to_language
+from typing import List
+
+
+
+########################################
+# Tasks
+########################################
+class FloresTranslationTask(Task):
+    DATASET_PATH = "facebook/flores"
+
+    def __init__(self, language_pair: str=None):
+        self.DATASET_NAME = self.language_pair = language_pair
+        self.src_code, self.tgt_code = language_pair.split("-")
+        self.src_lang = code_to_language(self.src_code[:3])
+        self.tgt_lang = code_to_language(self.tgt_code[:3])
+
+    def has_validation_docs(self):
+        return True
+
+    def has_test_docs(self):
+        return True
+
+    def has_training_docs(self):
+        return False
+
+    def test_docs(self):
+        return self.dataset["devtest"]
+
+    def validation_docs(self):
+        return self.dataset["dev"]
+
+    def doc_to_text(self, doc):
+        return f"{self.src_lang} phrase: " + doc[f"sentence_{self.src_lang}"]\
+            + f"\n{self.tgt_lang} phrase:"
+
+    def should_decontaminate(self):
+        return True
+
+    def doc_to_decontamination_query(self, doc):
+        return doc[f"sentence_{self.src_code}"]
+
+    def doc_to_target(self, doc):
+        return doc[f"sentence_{self.tgt_code}"]
+
+    def construct_requests(self, doc, ctx):
+        """Uses RequestFactory to construct Requests and returns an iterable of
+        Requests which will be sent to the LM.
+
+        :param doc:
+            The document as returned from training_docs, validation_docs, or test_docs.
+        :param ctx: str
+            The context string, generated by fewshot_context. This includes the natural
+            language description, as well as the few shot examples, and the question
+            part of the document for `doc`.
+        """
+        return rf.greedy_until(ctx, {"until": ["\n"]})
+
+    def process_results(self, doc, results):
+        # These metrics are corpus-level not sentence level, so we'll hide the
+        # results in this dict and compute the corpus score in the aggregate method
+        ref_pred = (self.doc_to_target(doc), results)
+        return {
+            "bleu": ref_pred,
+            "chrf": ref_pred,
+            "ter": ref_pred,
+        }
+
+    def aggregation(self):
+        """
+        :returns: {str: [float] -> float}
+            A dictionary where keys are the names of submetrics and values are
+            functions that aggregate a list of metrics
+        """
+        return {
+            "bleu": metrics.bleu,
+            "chrf": metrics.chrf,
+            "ter": metrics.ter,
+        }
+
+    def higher_is_better(self):
+        """
+        :returns: {str: bool}
+            A dictionary where keys are the names of submetrics and values are
+            whether a higher value of the submetric is better
+        """
+        return {
+            "bleu": True,
+            "chrf": True,
+            "ter": False,
+        }
+
+    def __str__(self):
+        return f"Flores200 {self.src_lang} to {self.tgt_lang} Task"
+
+def create_translation_task(language_pair, version=0):
+    class TranslationTask(FloresTranslationTask):
+        VERSION = version
+
+        def __init__(self):
+            super().__init__(language_pair)
+
+    return TranslationTask
+
+def create_tasks_from_list(lang_list):
+    """Symmetrically create all flores200 tasks from a list of languages.
+    """
+    return {f"flores200-{src}-{tgt}":create_translation_task(f"{src}-{tgt}")
+            for src, tgt in permutations(lang_list, 2)}