add XX->en direction to translation tasks

lm_eval/tasks/translation/iwslt2017_ar-en.yaml

 # Generated by utils.py
-dataset_name: iwslt2017-ar-en
+dataset_name: iwslt2017-en-ar
 dataset_path: iwslt2017
 doc_to_target: ' {{translation["en"]}}'
 doc_to_text: 'Arabic phrase: {{translation["ar"]}}

lm_eval/tasks/translation/utils.py

@@ -30,7 +30,7 @@ gpt3_translation_benchmarks = {
 LANGUAGES = {
     **gpt3_translation_benchmarks,
     # "wmt20": sacrebleu.get_langpairs_for_testset("wmt20"),
-    "iwslt2017": ["en-ar", "ar-en"],  # Arabic
+    "iwslt2017": ["en-ar"],  # Arabic
 }
 
 
@@ -49,45 +49,48 @@ def gen_lang_yamls(output_dir: str, overwrite: bool) -> None:
     """
     err = []
     for lang in LANGUAGES.keys():
-        for lang_pair in LANGUAGES[lang]:
-            file_name = f"{lang}_{lang_pair}.yaml"
-            try:
-                src_lang, _, tgt_lang = lang_pair.partition("-")
-                source, target = code_to_language(src_lang), code_to_language(tgt_lang)
-
-                groups = ["greedy_until", "translation", lang]
-                if lang in gpt3_translation_benchmarks.keys():
-                    groups += ["gpt3_translation_benchmarks"]
-
-                with open(
-                    f"{output_dir}/{file_name}",
-                    "w" if overwrite else "x",
-                    encoding="utf8",
-                ) as f:
-                    f.write("# Generated by utils.py\n")
-                    yaml.dump(
-                        {
-                            "include": "wmt_common_yaml",
-                            "group": groups,
-                            "dataset_path": lang,
-                            "dataset_name": lang_pair
-                            if not (lang == "iwslt2017")
-                            else "iwslt2017-" + lang_pair,
-                            "task": f"{lang}-{lang_pair}",
-                            "doc_to_text": f"{source} phrase: "
-                            + "{{translation["
-                            + f'"{src_lang}"'
-                            + "]}}\n"
-                            + f"{target} phrase:",
-                            "doc_to_target": " {{"
-                            + "translation["
-                            + f'"{tgt_lang}"]'
-                            + "}}",
-                        },
-                        f,
-                    )
-            except FileExistsError:
-                err.append(file_name)
+        for dataset_name in LANGUAGES[lang]:
+            src_lang, _, tgt_lang = dataset_name.partition("-")
+            for src, tgt in [[src_lang, tgt_lang], [tgt_lang, src_lang]]:
+                # both translation directions for each lang pair
+                lang_pair = src + "-" + tgt
+                file_name = f"{lang}_{lang_pair}.yaml"
+                try:
+                    source, target = code_to_language(src), code_to_language(tgt)
+
+                    groups = ["greedy_until", "translation", lang]
+                    if lang in gpt3_translation_benchmarks.keys():
+                        groups += ["gpt3_translation_benchmarks"]
+
+                    with open(
+                        f"{output_dir}/{file_name}",
+                        "w" if overwrite else "x",
+                        encoding="utf8",
+                    ) as f:
+                        f.write("# Generated by utils.py\n")
+                        yaml.dump(
+                            {
+                                "include": "wmt_common_yaml",
+                                "group": groups,
+                                "dataset_path": lang,
+                                "dataset_name": dataset_name
+                                if not (lang == "iwslt2017")
+                                else "iwslt2017-" + dataset_name,
+                                "task": f"{lang}-{lang_pair}",
+                                "doc_to_text": f"{source} phrase: "
+                                + "{{translation["
+                                + f'"{src}"'
+                                + "]}}\n"
+                                + f"{target} phrase:",
+                                "doc_to_target": " {{"
+                                + "translation["
+                                + f'"{tgt}"]'
+                                + "}}",
+                            },
+                            f,
+                        )
+                except FileExistsError:
+                    err.append(file_name)
 
     if len(err) > 0:
         raise FileExistsError(

lm_eval/tasks/translation/wmt14_en-fr.yaml

+# Generated by utils.py
+dataset_name: fr-en
+dataset_path: wmt14
+doc_to_target: ' {{translation["fr"]}}'
+doc_to_text: 'English phrase: {{translation["en"]}}
+
+  French phrase:'
+group:
+- greedy_until
+- translation
+- wmt14
+- gpt3_translation_benchmarks
+include: wmt_common_yaml
+task: wmt14-en-fr

lm_eval/tasks/translation/wmt16_en-de.yaml

+# Generated by utils.py
+dataset_name: de-en
+dataset_path: wmt16
+doc_to_target: ' {{translation["de"]}}'
+doc_to_text: 'English phrase: {{translation["en"]}}
+
+  German phrase:'
+group:
+- greedy_until
+- translation
+- wmt16
+- gpt3_translation_benchmarks
+include: wmt_common_yaml
+task: wmt16-en-de

lm_eval/tasks/translation/wmt16_en-ro.yaml

+# Generated by utils.py
+dataset_name: ro-en
+dataset_path: wmt16
+doc_to_target: ' {{translation["ro"]}}'
+doc_to_text: 'English phrase: {{translation["en"]}}
+
+  Romanian phrase:'
+group:
+- greedy_until
+- translation
+- wmt16
+- gpt3_translation_benchmarks
+include: wmt_common_yaml
+task: wmt16-en-ro