-
Notifications
You must be signed in to change notification settings - Fork 370
Adding a range of multilingual evals #832
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Draft
clefourrier
wants to merge
15
commits into
main
Choose a base branch
from
clem_mmlupro
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Draft
Changes from 6 commits
Commits
Show all changes
15 commits
Select commit
Hold shift + click to select a range
5b3cd26
added instruct specific evals
clefourrier 75af52d
tmp
clefourrier ffa43cc
fix
clefourrier a94973a
fix
clefourrier 2852db8
tmp
clefourrier a549107
focused data on actually existing languages
clefourrier 031e09c
Fix GPQA and index extractive metric (#829)
clefourrier bc5a564
Complete TranslationLiterals for Language.ESTONIAN (#779)
spyysalo bae9544
belebele split
clefourrier 78a5c23
updated prompts/instruction management
clefourrier 91ad573
tmp
clefourrier 6f5d92f
added enforce eager
clefourrier 2a10826
small prompt fix
clefourrier 2114c61
one file per eval
clefourrier ee6149e
adding mgsm
clefourrier File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,47 @@ | ||
| extended|belebele_instruct_deu_Latn|0|0 | ||
| extended|belebele_instruct_fra_Latn|0|0 | ||
| extended|belebele_instruct_ita_Latn|0|0 | ||
| extended|belebele_instruct_por_Latn|0|0 | ||
| extended|belebele_instruct_spa_Latn|0|0 | ||
| extended|global_mmlu_instruct_amh|0|0 | ||
| extended|global_mmlu_instruct_ara|0|0 | ||
| extended|global_mmlu_instruct_ben|0|0 | ||
| extended|global_mmlu_instruct_ces|0|0 | ||
| extended|global_mmlu_instruct_deu|0|0 | ||
| extended|global_mmlu_instruct_ell|0|0 | ||
| extended|global_mmlu_instruct_eng|0|0 | ||
| extended|global_mmlu_instruct_spa|0|0 | ||
| extended|global_mmlu_instruct_fas|0|0 | ||
| extended|global_mmlu_instruct_fra|0|0 | ||
| extended|global_mmlu_instruct_hau|0|0 | ||
| extended|global_mmlu_instruct_heb|0|0 | ||
| extended|global_mmlu_instruct_hin|0|0 | ||
| extended|global_mmlu_instruct_ind|0|0 | ||
| extended|global_mmlu_instruct_ibo|0|0 | ||
| extended|global_mmlu_instruct_ita|0|0 | ||
| extended|global_mmlu_instruct_jpn|0|0 | ||
| extended|global_mmlu_instruct_kor|0|0 | ||
| extended|global_mmlu_instruct_kir|0|0 | ||
| extended|global_mmlu_instruct_lit|0|0 | ||
| extended|global_mmlu_instruct_mlg|0|0 | ||
| extended|global_mmlu_instruct_msa|0|0 | ||
| extended|global_mmlu_instruct_nep|0|0 | ||
| extended|global_mmlu_instruct_nld|0|0 | ||
| extended|global_mmlu_instruct_nor|0|0 | ||
| extended|global_mmlu_instruct_pol|0|0 | ||
| extended|global_mmlu_instruct_por|0|0 | ||
| extended|global_mmlu_instruct_ron|0|0 | ||
| extended|global_mmlu_instruct_rus|0|0 | ||
| extended|global_mmlu_instruct_sin|0|0 | ||
| extended|global_mmlu_instruct_sna|0|0 | ||
| extended|global_mmlu_instruct_som|0|0 | ||
| extended|global_mmlu_instruct_srp|0|0 | ||
| extended|global_mmlu_instruct_swe|0|0 | ||
| extended|global_mmlu_instruct_swa|0|0 | ||
| extended|global_mmlu_instruct_tel|0|0 | ||
| extended|global_mmlu_instruct_tur|0|0 | ||
| extended|global_mmlu_instruct_ukr|0|0 | ||
| extended|global_mmlu_instruct_vie|0|0 | ||
| extended|global_mmlu_instruct_yor|0|0 | ||
| extended|global_mmlu_instruct_zho|0|0 | ||
| extended|mmlu_pro|0|0 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,293 @@ | ||
| # MIT License | ||
|
|
||
| # Copyright (c) 2024 The HuggingFace Team | ||
|
|
||
| # Permission is hereby granted, free of charge, to any person obtaining a copy | ||
| # of this software and associated documentation files (the "Software"), to deal | ||
| # in the Software without restriction, including without limitation the rights | ||
| # to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | ||
| # copies of the Software, and to permit persons to whom the Software is | ||
| # furnished to do so, subject to the following conditions: | ||
|
|
||
| # The above copyright notice and this permission notice shall be included in all | ||
| # copies or substantial portions of the Software. | ||
|
|
||
| # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
| # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
| # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | ||
| # AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
| # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | ||
| # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | ||
| # SOFTWARE. | ||
|
|
||
|
|
||
| import numpy as np | ||
|
|
||
| from lighteval.metrics.dynamic_metrics import ( | ||
| IndicesExtractionConfig, | ||
| multilingual_extractive_match_metric, | ||
| ) | ||
| from lighteval.metrics.metrics import MetricCategory, MetricUseCase, SampleLevelMetric | ||
| from lighteval.metrics.metrics_sample import ( | ||
| PassAtK, | ||
| ) | ||
| from lighteval.tasks.default_prompts import LETTER_INDICES | ||
| from lighteval.tasks.lighteval_task import LightevalTaskConfig | ||
| from lighteval.tasks.requests import Doc | ||
| from lighteval.utils.language import Language | ||
|
|
||
|
|
||
| TASKS_TABLE = [] | ||
|
|
||
| lang_to_literal = { | ||
| "deu": Language.GERMAN, | ||
| "fra": Language.FRENCH, | ||
| "ita": Language.ITALIAN, | ||
| "por": Language.PORTUGUESE, | ||
| "spa": Language.SPANISH, | ||
| } | ||
|
|
||
|
|
||
| def belebele_prompt(line, task_name: str = None): | ||
| lang_to_template = { | ||
| "eng_Latn": "Given the following passage, query, and answer choices, output the letter corresponding to the correct answer. The last line of your response should be of the following format: 'Answer: $LETTER' (without quotes) where LETTER is one of A, B, C, or D. Think step by step before answering.\n\n###\nPassage:\n{Passage}\n###\nQuery:\n{Question}\n###\nChoices:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "deu_Latn": "Gib basierend auf dem folgenden Textabschnitt, der Frage und den Antwortmöglichkeiten den Buchstaben aus, der der richtigen Antwort entspricht. Die letzte Zeile deiner Antwort sollte folgendes Format haben: 'Antwort: $BUCHSTABE' (ohne Anführungszeichen), wobei BUCHSTABE einer der folgenden ist: A, B, C oder D. Denke Schritt für Schritt, bevor du antwortest.\n\n###\nTextabschnitt:\n{Passage}\n###\nFrage:\n{Question}\n###\nAntwortmöglichkeiten:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "fra_Latn": "A partir du passage suivant, de la question et des choix de réponses, indiquez la lettre correspondant à la bonne réponse. La dernière ligne de votre réponse doit avoir le format suivant : 'Réponse: '$LETTRE' (sans les guillemets) où LETTRE est l'une des lettres: A, B, C ou D. Réfléchissez étape par étape avant de répondre.\n\n###\nPassage:\n{Passage}\n###\nRequête:\n{Question}\n###\nChoix:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "ita_Latn": "Dato il seguente passaggio, un quesito e le diverse opzioni per una risposta, indicare la lettera corrispondente alla risposta corretta. L'ultima riga della risposta deve avere il seguente formato: 'Risposta: $LETTERA' (senza virgolette), e LETTERA è necessariamente una tra A, B, C, D. Prima di rispondere, è importante che si ragioni passo per passo.\n\n###\nPassaggio:\n{Passage}\n###\nQuesito:\n{Question}\n###\nOpzioni:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "por_Latn": "Tendo em conta a seguinte passagem, pergunta e opções de resposta, indique a letra correspondente à resposta correta. A última linha da sua resposta deve ter o seguinte formato: 'Resposta: $LETRA' (sem aspas) em que LETRA é uma de A, B, C ou D. Pense passo a passo antes de responder.\n\n###\nPassagem:\n{Passage}\n###\nPergunta:\n{Question}\n###\nOpções:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "spa_Latn": "Dado el siguiente contexto, pregunta y opciones para la respuesta, escriba la letra correspondiente a la respuesta correcta. La última línea de su respuesta debe seguir el siguiente formato: 'Respuesta: $LETTER' (sin comillas) donde LETTER es A, B, C o D. Piense paso a paso antes de responder.\n\n###\nContexto:\n{Passage}\n###\nPregunta:\n{Question}\n###\nOpciones:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| } | ||
|
|
||
| gold_index = int(line["correct_answer_num"]) - 1 | ||
| choices = [line["mc_answer1"], line["mc_answer2"], line["mc_answer3"], line["mc_answer4"]] | ||
| query_template = lang_to_template.get(line["dialect"], "eng_Latn") | ||
| query = query_template.format( | ||
| A=choices[0], | ||
| B=choices[1], | ||
| C=choices[2], | ||
| D=choices[3], | ||
| Passage=line["flores_passage"], | ||
| Question=line["question"], | ||
| ) | ||
| instruction = query_template.split("\n\n###")[0] | ||
|
|
||
| return Doc( | ||
| task_name=task_name, | ||
| query=query, | ||
| choices=LETTER_INDICES[: len(choices)], | ||
| gold_index=gold_index, | ||
| instruction=instruction, | ||
| ) | ||
|
|
||
|
|
||
| BELEBELE_TASKS = [ | ||
| LightevalTaskConfig( | ||
| name=f"belebele_instruct_{lang}_Latn", | ||
| prompt_function=belebele_prompt, | ||
| suite=["extended"], | ||
| hf_repo="facebook/belebele", | ||
| hf_subset=f"{lang}_Latn", | ||
| evaluation_splits=["test"], | ||
| hf_avail_splits=["test"], | ||
| few_shots_split=None, | ||
| few_shots_select=None, | ||
| generation_size=32768, # needed for reasoning models like R1 | ||
| metric=[ | ||
| SampleLevelMetric( | ||
| metric_name="pass@1:1_samples", | ||
| sample_level_fn=PassAtK( | ||
| k=1, | ||
| n=1, | ||
| sample_scoring_function=lambda pred, ref, doc: multilingual_extractive_match_metric( | ||
| language=lang_to_literal[lang], | ||
| gold_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")], | ||
| pred_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")], | ||
| precision=6, | ||
| ).sample_level_fn([ref], [pred], doc), | ||
| ).compute, | ||
| category=MetricCategory.GENERATIVE_SAMPLING, | ||
| use_case=MetricUseCase.REASONING, | ||
| corpus_level_fn=np.mean, | ||
| higher_is_better=True, | ||
| ) | ||
| ], | ||
| stop_sequence=[], # no stop sequence, will use eos token | ||
| trust_dataset=True, | ||
| version=1, | ||
| ) | ||
| for lang in [ | ||
| "deu", | ||
| "fra", | ||
| "ita", | ||
| "por", | ||
| "spa", | ||
| ] | ||
| ] | ||
| TASKS_TABLE.extend(BELEBELE_TASKS) | ||
|
|
||
|
|
||
| class GlobalMMLUPrompt: | ||
| def __init__(self, lang): | ||
| self.lang = lang | ||
| self.lang_to_template = { | ||
| "eng": "Given the following query and answer choices, output the letter corresponding to the correct answer. The last line of your response should be of the following format: 'Answer: $LETTER' (without quotes) where LETTER is one of A, B, C, or D. Think step by step before answering.\n\n###\nQuery:\n{Question}\n###\nChoices:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "deu": "Gib basierend auf der folgenden Frage und den Antwortmöglichkeiten den Buchstaben aus, der der richtigen Antwort entspricht. Die letzte Zeile deiner Antwort sollte folgendes Format haben: 'Antwort: $BUCHSTABE' (ohne Anführungszeichen), wobei BUCHSTABE einer der folgenden ist: A, B, C oder D. Denke Schritt für Schritt, bevor du antwortest.\n\n###\nFrage:\n{Question}\n###\nAntwortmöglichkeiten:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "fra": "A partir de la question et des choix de réponses suivants, indiquez la lettre correspondant à la bonne réponse. La dernière ligne de votre réponse doit avoir le format suivant : 'Réponse: '$LETTRE' (sans les guillemets) où LETTRE est l'une des lettres: A, B, C ou D. Réfléchissez étape par étape avant de répondre.\n\n###\nRequête:\n{Question}\n###\nChoix:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "ita": "Dato il seguente quesito e le diverse opzioni per una risposta, indicare la lettera corrispondente alla risposta corretta. L'ultima riga della risposta deve avere il seguente formato: 'Risposta: $LETTERA' (senza virgolette), e LETTERA è necessariamente una tra A, B, C, D. Prima di rispondere, è importante che si ragioni passo per passo.\n\n###\nQuesito:\n{Question}\n###\nOpzioni:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "por": "Tendo em conta a seguinte pergunta e opções de resposta, indique a letra correspondente à resposta correta. A última linha da sua resposta deve ter o seguinte formato: 'Resposta: $LETRA' (sem aspas) em que LETRA é uma de A, B, C ou D. Pense passo a passo antes de responder.\n\n###\nPergunta:\n{Question}\n###\nOpções:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| "spa": "Dado el siguiente pregunta y opciones para la respuesta, escriba la letra correspondiente a la respuesta correcta. La última línea de su respuesta debe seguir el siguiente formato: 'Respuesta: $LETTER' (sin comillas) donde LETTER es A, B, C o D. Piense paso a paso antes de responder.\n\\###\nPregunta:\n{Question}\n###\nOpciones:\nA) {A}\nB) {B}\nC) {C}\nD) {D}", | ||
| } | ||
|
|
||
| def prompt(self, line, task_name: str = None): | ||
| gold_index = LETTER_INDICES.index(line["answer"]) | ||
| choices = [line["option_a"], line["option_b"], line["option_c"], line["option_d"]] | ||
| query_template = self.lang_to_template.get(self.lang, "eng") | ||
| query = query_template.format( | ||
| A=choices[0], | ||
| B=choices[1], | ||
| C=choices[2], | ||
| D=choices[3], | ||
| Question=line["question"], | ||
| ) | ||
| instruction = query_template.split("\n\n###")[0] | ||
|
|
||
| return Doc( | ||
| task_name=task_name, | ||
| query=query, | ||
| choices=LETTER_INDICES[: len(choices)], | ||
| gold_index=gold_index, | ||
| instruction=instruction, | ||
| ) | ||
|
|
||
|
|
||
| GLOBAL_MMLU_TASKS = [ | ||
| LightevalTaskConfig( | ||
| name=f"global_mmlu_instruct_{language.value}", | ||
| prompt_function=GlobalMMLUPrompt(language.value).prompt, | ||
| suite=["extended"], | ||
| hf_repo="CohereForAI/Global-MMLU", | ||
| hf_subset=lang, | ||
| evaluation_splits=("test",), | ||
| few_shots_split="dev", | ||
| metric=[ | ||
| SampleLevelMetric( | ||
| metric_name="pass@1:1_samples", | ||
| sample_level_fn=PassAtK( | ||
| k=1, | ||
| n=1, | ||
| sample_scoring_function=lambda pred, ref, doc: multilingual_extractive_match_metric( | ||
| language=language, | ||
| gold_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")], | ||
| pred_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")], | ||
| precision=6, | ||
| ).sample_level_fn([ref], [pred], doc), | ||
| ).compute, | ||
| category=MetricCategory.GENERATIVE_SAMPLING, | ||
| use_case=MetricUseCase.REASONING, | ||
| corpus_level_fn=np.mean, | ||
| higher_is_better=True, | ||
| ) | ||
| ], | ||
| generation_size=32768, # needed for reasoning models like R1 | ||
| stop_sequence=[], # no stop sequence, will use eos token | ||
| ) | ||
| for lang, language in [ | ||
| ("am", Language.AMHARIC), | ||
| ("ar", Language.ARABIC), | ||
| ("bn", Language.BENGALI), | ||
| ("cs", Language.CZECH), | ||
| ("de", Language.GERMAN), | ||
| ("el", Language.GREEK), | ||
| ("en", Language.ENGLISH), | ||
| ("es", Language.SPANISH), | ||
| ("fa", Language.PERSIAN), | ||
| # ("fil", Language.FILIPINO), | ||
| ("fr", Language.FRENCH), | ||
| ("ha", Language.HAUSA), | ||
| ("he", Language.HEBREW), | ||
| ("hi", Language.HINDI), | ||
| ("id", Language.INDONESIAN), | ||
| ("ig", Language.IGBO), | ||
| ("it", Language.ITALIAN), | ||
| ("ja", Language.JAPANESE), | ||
| ("ko", Language.KOREAN), | ||
| ("ky", Language.KYRGYZ), | ||
| ("lt", Language.LITHUANIAN), | ||
| ("mg", Language.MALAGASY), | ||
| ("ms", Language.MALAY), | ||
| ("ne", Language.NEPALI), | ||
| ("nl", Language.DUTCH), | ||
| ("ny", Language.NORWEGIAN), | ||
| ("pl", Language.POLISH), | ||
| ("pt", Language.PORTUGUESE), | ||
| ("ro", Language.ROMANIAN), | ||
| ("ru", Language.RUSSIAN), | ||
| ("si", Language.SINHALA), | ||
| ("sn", Language.SHONA), | ||
| ("so", Language.SOMALI), | ||
| ("sr", Language.SERBIAN), | ||
| ("sv", Language.SWEDISH), | ||
| ("sw", Language.SWAHILI), | ||
| ("te", Language.TELUGU), | ||
| ("tr", Language.TURKISH), | ||
| ("uk", Language.UKRAINIAN), | ||
| ("vi", Language.VIETNAMESE), | ||
| ("yo", Language.YORUBA), | ||
| ("zh", Language.CHINESE), | ||
| ] | ||
| ] | ||
| TASKS_TABLE.extend(GLOBAL_MMLU_TASKS) | ||
|
|
||
|
|
||
| def mmlu_pro(line, task_name: str = None): | ||
| num_choices = len(line["options"]) | ||
| instruction = f"Given the following question about {line['category']} and answer choices, output the letter corresponding to the correct answer. The last line of your response should be of the following format: 'Answer: $LETTER' (without quotes) where LETTER is one of {' ,'.join(LETTER_INDICES[: num_choices - 1])}, or {LETTER_INDICES[num_choices]}. Think step by step before answering.\n\n" | ||
| query = f"{instruction}###\nQuery:\n{line['question']}\n###\nChoices:" | ||
| query += "".join([f"\n{key}) {choice}" for key, choice in zip(LETTER_INDICES, line["options"])]) | ||
|
|
||
| return Doc( | ||
| task_name=task_name, | ||
| query=query, | ||
| choices=LETTER_INDICES[:num_choices], | ||
| gold_index=line["answer_index"], | ||
| instruction=instruction, | ||
| ) | ||
|
|
||
|
|
||
| mmlu_pro = LightevalTaskConfig( | ||
| name="mmlu_pro", | ||
| suite=["extended"], | ||
| prompt_function=mmlu_pro, | ||
| hf_repo="TIGER-Lab/MMLU-Pro", | ||
| hf_subset="default", | ||
| hf_avail_splits=["validation", "test"], | ||
| evaluation_splits=["test"], | ||
| few_shots_split="validation", | ||
| few_shots_select=None, | ||
| generation_size=32768, # needed for reasoning models like R1 | ||
| stop_sequence=[], # no stop sequence, will use eos token | ||
| metric=[ | ||
| SampleLevelMetric( | ||
| metric_name="pass@1:1_samples", | ||
| sample_level_fn=PassAtK( | ||
| k=1, | ||
| n=1, | ||
| sample_scoring_function=lambda pred, ref, doc: multilingual_extractive_match_metric( | ||
| language=Language.ENGLISH, | ||
| gold_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")], | ||
| pred_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")], | ||
| precision=6, | ||
| ).sample_level_fn([ref], [pred], doc), | ||
| ).compute, | ||
| category=MetricCategory.GENERATIVE_SAMPLING, | ||
| use_case=MetricUseCase.REASONING, | ||
| corpus_level_fn=np.mean, | ||
| higher_is_better=True, | ||
| ) | ||
| ], | ||
| trust_dataset=True, | ||
| version=0, | ||
| ) | ||
|
|
||
| TASKS_TABLE.append(mmlu_pro) | ||
| print(TASKS_TABLE) | ||
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe we should call this
belebele_instruct_5_{lang}_Latnorbelebele_instruct_smollm_{lang}_Latnto distinguish from the general case with more languages?The alternative would be to have a separate
belebele_instruct_en_{lang}_{script}for the full set of languages, but with English instructionsUh oh!
There was an error while loading. Please reload this page.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Will add the latter, and have a
`belebele_native_inst_{lang}" vs "belebele_en_inst_{lang}"
:)