LLM Benchmarks#
Below is the list of supported LLM benchmarks. Click on a benchmark name to jump to details.
Benchmark Name |
Pretty Name |
Task Categories |
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Benchmark Details#
AA-LCR#
Dataset Name:
aa_lcrDataset ID: evalscope/AA-LCR
Description:
AA-LCR (Artificial Analysis Long Context Retrieval) is a benchmark for evaluating long-context retrieval and reasoning capabilities of language models across multiple documents.
Task Categories:
Knowledge,LongContext,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
defaultExtra Parameters:
{
"text_dir": {
"type": "str | null",
"description": "Local directory containing extracted AA-LCR text files; if null will auto-download & extract.",
"value": null
}
}
Prompt Template:
View
BEGIN INPUT DOCUMENTS
{documents_text}
END INPUT DOCUMENTS
Answer the following question using the input documents provided above.
START QUESTION
{question}
END QUESTION
AIME-2024#
Dataset Name:
aime24Dataset ID: HuggingFaceH4/aime_2024
Description:
The AIME 2024 benchmark is based on problems from the American Invitational Mathematics Examination, a prestigious high school mathematics competition. This benchmark tests a model’s ability to solve challenging mathematics problems by generating step-by-step solutions and providing the correct final answer.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
AIME-2025#
Dataset Name:
aime25Dataset ID: opencompass/AIME2025
Description:
The AIME 2025 benchmark is based on problems from the American Invitational Mathematics Examination, a prestigious high school mathematics competition. This benchmark tests a model’s ability to solve challenging mathematics problems by generating step-by-step solutions and providing the correct final answer.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
AIME2025-II,AIME2025-IPrompt Template:
View
Solve the following math problem step by step. Put your answer inside \boxed{{}}.
{question}
Remember to put your answer inside \boxed{{}}.
AlpacaEval2.0#
Dataset Name:
alpaca_evalDataset ID: AI-ModelScope/alpaca_eval
Description:
Alpaca Eval 2.0 is an enhanced framework for evaluating instruction-following language models, featuring an improved auto-annotator, updated baselines, and continuous preference calculation to provide more accurate and cost-effective model assessments. Currently not support
length-controlled winrate; the official Judge model isgpt-4-1106-preview, while the baseline model isgpt-4-turbo.Task Categories:
Arena,InstructionFollowingEvaluation Metrics:
winrateAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
alpaca_eval_gpt4_baselinePrompt Template:
View
{question}
AMC#
Dataset Name:
amcDataset ID: evalscope/amc_22-24
Description:
AMC (American Mathematics Competitions) is a series of mathematics competitions for high school students.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
amc22,amc23,amc24Prompt Template:
View
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
ARC#
Dataset Name:
arcDataset ID: allenai/ai2_arc
Description:
The ARC (AI2 Reasoning Challenge) benchmark is designed to evaluate the reasoning capabilities of AI models through multiple-choice questions derived from science exams. It includes two subsets: ARC-Easy and ARC-Challenge, which vary in difficulty.
Task Categories:
MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
ARC-Challenge,ARC-EasyPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
ArenaHard#
Dataset Name:
arena_hardDataset ID: AI-ModelScope/arena-hard-auto-v0.1
Description:
ArenaHard is a benchmark designed to evaluate the performance of large language models in a competitive setting, where models are pitted against each other in a series of tasks to determine their relative strengths and weaknesses. It includes a set of challenging tasks that require reasoning, understanding, and generation capabilities. Currently not support
style-controlled winrate; the official Judge model isgpt-4-1106-preview, while the baseline model isgpt-4-0314.Task Categories:
Arena,InstructionFollowingEvaluation Metrics:
winrateAggregation Methods:
eloRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
{question}
BBH#
Dataset Name:
bbhDataset ID: evalscope/bbh
Description:
The BBH (Big Bench Hard) benchmark is a collection of challenging tasks designed to evaluate the reasoning capabilities of AI models. It includes both free-form and multiple-choice tasks, covering a wide range of reasoning skills.
Task Categories:
ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 3-shot
Subsets:
boolean_expressions,causal_judgement,date_understanding,disambiguation_qa,dyck_languages,formal_fallacies,geometric_shapes,hyperbaton,logical_deduction_five_objects,logical_deduction_seven_objects,logical_deduction_three_objects,movie_recommendation,multistep_arithmetic_two,navigate,object_counting,penguins_in_a_table,reasoning_about_colored_objects,ruin_names,salient_translation_error_detection,snarks,sports_understanding,temporal_sequences,tracking_shuffled_objects_five_objects,tracking_shuffled_objects_seven_objects,tracking_shuffled_objects_three_objects,web_of_lies,word_sortingPrompt Template:
View
Q: {question}
A: Let's think step by step. Put your final answer in the format of "So the answer is [ANSWER]" (without quotes and markdown) where [ANSWER] is the answer to the problem.
BioMixQA#
Dataset Name:
biomix_qaDataset ID: extraordinarylab/biomix-qa
Description:
BiomixQA is a curated biomedical question-answering dataset. BiomixQA has been utilized to validate the Knowledge Graph based Retrieval-Augmented Generation (KG-RAG) framework across different LLMs.
Task Categories:
Knowledge,MCQ,MedicalEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
BroadTwitterCorpus#
Dataset Name:
broad_twitter_corpusDataset ID: extraordinarylab/broad-twitter-corpus
Description:
BroadTwitterCorpus is a dataset of tweets collected over stratified times, places and social uses. The goal is to represent a broad range of activities, giving a dataset more representative of the language used in this hardest of social media formats to process.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
C-Eval#
Dataset Name:
cevalDataset ID: evalscope/ceval
Description:
C-Eval is a benchmark designed to evaluate the performance of AI models on Chinese exams across various subjects, including STEM, social sciences, and humanities. It consists of multiple-choice questions that test knowledge and reasoning abilities in these areas.
Task Categories:
Chinese,Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
accountant,advanced_mathematics,art_studies,basic_medicine,business_administration,chinese_language_and_literature,civil_servant,clinical_medicine,college_chemistry,college_economics,college_physics,college_programming,computer_architecture,computer_network,discrete_mathematics,education_science,electrical_engineer,environmental_impact_assessment_engineer,fire_engineer,high_school_biology,high_school_chemistry,high_school_chinese,high_school_geography,high_school_history,high_school_mathematics,high_school_physics,high_school_politics,ideological_and_moral_cultivation,law,legal_professional,logic,mao_zedong_thought,marxism,metrology_engineer,middle_school_biology,middle_school_chemistry,middle_school_geography,middle_school_history,middle_school_mathematics,middle_school_physics,middle_school_politics,modern_chinese_history,operating_system,physician,plant_protection,probability_and_statistics,professional_tour_guide,sports_science,tax_accountant,teacher_qualification,urban_and_rural_planner,veterinary_medicinePrompt Template:
View
以下是中国关于{subject}的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：[LETTER]"（不带引号），其中 [LETTER] 是 A、B、C、D 中的一个。
问题：{question}
选项：
{choices}
Chinese-SimpleQA#
Dataset Name:
chinese_simpleqaDataset ID: AI-ModelScope/Chinese-SimpleQA
Description:
Chinese SimpleQA is a Chinese question-answering dataset designed to evaluate the performance of language models on simple factual questions. It includes a variety of topics and is structured to test the model’s ability to understand and generate correct answers in Chinese.
Task Categories:
Chinese,Knowledge,QAEvaluation Metrics:
is_correct,is_incorrect,is_not_attemptedAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
中华文化,人文与社会科学,工程、技术与应用科学,生活、艺术与文化,社会,自然与自然科学Prompt Template:
View
请回答问题：
{question}
C-MMLU#
Dataset Name:
cmmluDataset ID: evalscope/cmmlu
Description:
C-MMLU is a benchmark designed to evaluate the performance of AI models on Chinese language tasks, including reading comprehension, text classification, and more.
Task Categories:
Chinese,Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
agronomy,anatomy,ancient_chinese,arts,astronomy,business_ethics,chinese_civil_service_exam,chinese_driving_rule,chinese_food_culture,chinese_foreign_policy,chinese_history,chinese_literature,chinese_teacher_qualification,clinical_knowledge,college_actuarial_science,college_education,college_engineering_hydrology,college_law,college_mathematics,college_medical_statistics,college_medicine,computer_science,computer_security,conceptual_physics,construction_project_management,economics,education,electrical_engineering,elementary_chinese,elementary_commonsense,elementary_information_and_technology,elementary_mathematics,ethnology,food_science,genetics,global_facts,high_school_biology,high_school_chemistry,high_school_geography,high_school_mathematics,high_school_physics,high_school_politics,human_sexuality,international_law,journalism,jurisprudence,legal_and_moral_basis,logical,machine_learning,management,marketing,marxist_theory,modern_chinese,nutrition,philosophy,professional_accounting,professional_law,professional_medicine,professional_psychology,public_relations,security_study,sociology,sports_science,traditional_chinese_medicine,virology,world_history,world_religionsPrompt Template:
View
回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：[LETTER]"（不带引号），其中 [LETTER] 是 {letters} 中的一个。请在回答前进行一步步思考。
问题：{question}
选项：
{choices}
CoinFlip#
Dataset Name:
coin_flipDataset ID: extraordinarylab/coin-flip
Description:
CoinFlip is a symbolic reasoning dataset that tests an LLM’s ability to track binary state changes through a sequence of actions. Each example describes whether a coin is flipped or not by different person, requiring logical inference to determine the final state (heads or tails).
Task Categories:
Reasoning,Yes/NoEvaluation Metrics:
accuracy,f1_score,precision,recall,yes_ratioAggregation Methods:
f1Requires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Solve the following coin flip problem step by step. The last line of your response should be of the form "ANSWER: [ANSWER]" (without quotes) where [ANSWER] is the answer to the problem.
{question}
Remember to put your answer on its own line at the end in the form "ANSWER: [ANSWER]" (without quotes) where [ANSWER] is the answer YES or NO to the problem.
Reasoning:
CommonsenseQA#
Dataset Name:
commonsense_qaDataset ID: extraordinarylab/commonsense-qa
Description:
CommonsenseQA requires different types of commonsense knowledge to predict the correct answers.
Task Categories:
Commonsense,MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
Competition-MATH#
Dataset Name:
competition_mathDataset ID: evalscope/competition_math
Description:
The MATH (Mathematics) benchmark is designed to evaluate the mathematical reasoning abilities of AI models through a variety of problem types, including arithmetic, algebra, geometry, and more.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 4-shot
Subsets:
Level 1,Level 2,Level 3,Level 4,Level 5Prompt Template:
View
Problem:
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
CoNLL2003#
Dataset Name:
conll2003Dataset ID: evalscope/conll2003
Description:
The ConLL-2003 dataset is for the Named Entity Recognition (NER) task. It was introduced as part of the ConLL-2003 Shared Task conference and contains texts annotated with entities such as people, organizations, places, and various names.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
Copious#
Dataset Name:
copiousDataset ID: extraordinarylab/copious
Description:
Copious corpus is a gold standard corpus that covers a wide range of biodiversity entities, consisting of 668 documents downloaded from the Biodiversity Heritage Library with over 26K sentences and more than 28K entities.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
CrossNER#
Dataset Name:
cross_nerDataset ID: extraordinarylab/cross-ner
Description:
CrossNER is a fully-labelled collected of named entity recognition (NER) data spanning over five diverse domains (AI, Literature, Music, Politics, Science).
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
ai,literature,music,politics,sciencePrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
Data-Collection#
Dataset Name:
data_collectionDataset ID:
Description:
Custom Data collection, mixing multiple evaluation datasets for a unified evaluation, aiming to use less data to achieve a more comprehensive assessment of the model’s capabilities. Usage Reference
Task Categories:
CustomEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
default
DocMath#
Dataset Name:
docmathDataset ID: yale-nlp/DocMath-Eval
Description:
DocMath-Eval is a comprehensive benchmark focused on numerical reasoning within specialized domains. It requires the model to comprehend long and specialized documents and perform numerical reasoning to answer the given question.
Task Categories:
LongContext,Math,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
complong_testmini,compshort_testmini,simplong_testmini,simpshort_testminiPrompt Template:
View
Please read the following text and answer the question below.
<text>
{context}
</text>
{question}
Format your response as follows: "Therefore, the answer is (insert answer here)".
DrivelologyBinaryClassification#
Dataset Name:
drivel_binaryDataset ID: extraordinarylab/drivel-hub
Description:
Drivelology, a unique linguistic phenomenon characterised as “nonsense with depth” - utterances that are syntactically coherent yet pragmatically paradoxical, emotionally loaded, or rhetorically subversive.
Task Categories:
Yes/NoEvaluation Metrics:
accuracy,f1_score,precision,recall,yes_ratioAggregation Methods:
f1Requires LLM Judge: No
Default Shots: 0-shot
Subsets:
binary-classificationPrompt Template:
View
{question}
DrivelologyMultilabelClassification#
Dataset Name:
drivel_multilabelDataset ID: extraordinarylab/drivel-hub
Description:
Drivelology, a unique linguistic phenomenon characterised as “nonsense with depth” - utterances that are syntactically coherent yet pragmatically paradoxical, emotionally loaded, or rhetorically subversive.
Task Categories:
MCQEvaluation Metrics:
exact_match,f1_macro,f1_micro,f1_weightedAggregation Methods:
f1_weightedRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
multi-label-classificationPrompt Template:
View
{question}
DrivelologyNarrativeSelection#
Dataset Name:
drivel_selectionDataset ID: extraordinarylab/drivel-hub
Description:
Drivelology, a unique linguistic phenomenon characterised as “nonsense with depth” - utterances that are syntactically coherent yet pragmatically paradoxical, emotionally loaded, or rhetorically subversive.
Task Categories:
MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
multiple-choice-english-easy,multiple-choice-english-hardPrompt Template:
View
Tell me the best option in the following options which represents the underlying narrative of the text?
The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
DrivelologyNarrativeWriting#
Dataset Name:
drivel_writingDataset ID: extraordinarylab/drivel-hub
Description:
Drivelology, a unique linguistic phenomenon characterised as “nonsense with depth” - utterances that are syntactically coherent yet pragmatically paradoxical, emotionally loaded, or rhetorically subversive.
Task Categories:
Knowledge,ReasoningEvaluation Metrics:
bert_score,gpt_scoreAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
narrative-writing-englishPrompt Template:
View
You need to first read and understand the text given. Generate a detailed description to illustrate the implicit narrative of the text.
Please provide your response in English, with a clear and comprehensive explanation of the narrative.
Text: {text}
DROP#
Dataset Name:
dropDataset ID: AI-ModelScope/DROP
Description:
The DROP (Discrete Reasoning Over Paragraphs) benchmark is designed to evaluate the reading comprehension and reasoning capabilities of AI models. It includes a variety of tasks that require models to read passages and answer questions based on the content.
Task Categories:
ReasoningEvaluation Metrics:
em,f1Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 3-shot
Subsets:
defaultPrompt Template:
View
You will be asked to read a passage and answer a question. {drop_examples}
# Your Task
---
{query}
Think step by step, then write a line of the form "Answer: [ANSWER]" at the end of your response.
FRAMES#
Dataset Name:
framesDataset ID: iic/frames
Description:
FRAMES is a comprehensive evaluation dataset designed to test the capabilities of Retrieval-Augmented Generation (RAG) systems across factuality, retrieval accuracy, and reasoning.
Task Categories:
LongContext,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Please read the following text and answer the question below.
<text>
{context}
</text>
{question}
Format your response as follows: "Therefore, the answer is (insert answer here)".
GeneralArena#
Dataset Name:
general_arenaDataset ID: general_arena
Description:
GeneralArena is a custom benchmark designed to evaluate the performance of large language models in a competitive setting, where models are pitted against each other in custom tasks to determine their relative strengths and weaknesses. You should provide the model outputs in the format of a list of dictionaries, where each dictionary contains the model name and its report path. For detailed instructions on how to use this benchmark, please refer to the Arena User Guide.
Task Categories:
Arena,CustomEvaluation Metrics:
winrateAggregation Methods:
eloRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
defaultExtra Parameters:
{
"models": {
"type": "list[dict]",
"description": "List of model entries with name and report_path for arena comparison.",
"value": [
{
"name": "qwen-plus",
"report_path": "outputs/20250627_172550/reports/qwen-plus"
},
{
"name": "qwen2.5-7b",
"report_path": "outputs/20250627_172817/reports/qwen2.5-7b-instruct"
}
]
},
"baseline": {
"type": "str",
"description": "Baseline model name used for ELO and winrate comparisons.",
"value": "qwen2.5-7b"
}
}
System Prompt:
View
Please act as an impartial judge and evaluate the quality of the responses provided by two AI assistants to the user prompt displayed below. You will be given assistant A's answer and assistant B's answer. Your job is to evaluate which assistant's answer is better.
Begin your evaluation by generating your own answer to the prompt. You must provide your answers before judging any answers.
When evaluating the assistants' answers, compare both assistants' answers with your answer. You must identify and correct any mistakes or inaccurate information.
Then consider if the assistant's answers are helpful, relevant, and concise. Helpful means the answer correctly responds to the prompt or follows the instructions. Note when user prompt has any ambiguity or more than one interpretation, it is more helpful and appropriate to ask for clarifications or more information from the user than providing an answer based on assumptions. Relevant means all parts of the response closely connect or are appropriate to what is being asked. Concise means the response is clear and not verbose or excessive.
Then consider the creativity and novelty of the assistant's answers when needed. Finally, identify any missing important information in the assistants' answers that would be beneficial to include when responding to the user prompt.
After providing your explanation, you must output only one of the following choices as your final verdict with a label:
1. Assistant A is significantly better: [[A>>B]]
2. Assistant A is slightly better: [[A>B]]
3. Tie, relatively the same: [[A=B]]
4. Assistant B is slightly better: [[B>A]]
5. Assistant B is significantly better: [[B>>A]]
Example output: "My final verdict is tie: [[A=B]]".
View
<|User Prompt|>
{question}
<|The Start of Assistant A's Answer|>
{answer_1}
<|The End of Assistant A's Answer|>
<|The Start of Assistant B's Answer|>
{answer_2}
<|The End of Assistant B's Answer|>
General-MCQ#
Dataset Name:
general_mcqDataset ID: general_mcq
Description:
A general multiple-choice question answering dataset for custom evaluation. For detailed instructions on how to use this benchmark, please refer to the User Guide.
Task Categories:
Custom,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
回答下面的单项选择题，请选出其中的正确答案。你的回答的全部内容应该是这样的格式："答案：[LETTER]"（不带引号），其中 [LETTER] 是 {letters} 中的一个。
问题：{question}
选项：
{choices}
General-QA#
Dataset Name:
general_qaDataset ID: general_qa
Description:
A general question answering dataset for custom evaluation. For detailed instructions on how to use this benchmark, please refer to the User Guide.
Task Categories:
Custom,QAEvaluation Metrics:
BLEU,RougeAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
请回答问题
{question}
GeniaNER#
Dataset Name:
genia_nerDataset ID: extraordinarylab/genia-ner
Description:
GeniaNER consisting of 2,000 MEDLINE abstracts has been released with more than 400,000 words and almost 100,000 annotations for biological terms.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
GPQA-Diamond#
Dataset Name:
gpqa_diamondDataset ID: AI-ModelScope/gpqa_diamond
Description:
GPQA is a dataset for evaluating the reasoning ability of large language models (LLMs) on complex mathematical problems. It contains questions that require step-by-step reasoning to arrive at the correct answer.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
GSM8K#
Dataset Name:
gsm8kDataset ID: AI-ModelScope/gsm8k
Description:
GSM8K (Grade School Math 8K) is a dataset of grade school math problems, designed to evaluate the mathematical reasoning abilities of AI models.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 4-shot
Subsets:
mainPrompt Template:
View
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
HaluEval#
Dataset Name:
haluevalDataset ID: evalscope/HaluEval
Description:
HaluEval is a large collection of generated and human-annotated hallucinated samples for evaluating the performance of LLMs in recognizing hallucination.
Task Categories:
Hallucination,Knowledge,Yes/NoEvaluation Metrics:
accuracy,f1_score,precision,recall,yes_ratioAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
dialogue_samples,qa_samples,summarization_samplesPrompt Template:
View
{question}
HarveyNER#
Dataset Name:
harvey_nerDataset ID: extraordinarylab/harvey-ner
Description:
HarveyNER is a dataset with fine-grained locations annotated in tweets. This dataset presents unique challenges and characterizes many complex and long location mentions in informal descriptions.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
HealthBench#
Dataset Name:
health_benchDataset ID: openai-mirror/healthbench
Description:
HealthBench: a new benchmark designed to better measure capabilities of AI systems for health. Built in partnership with 262 physicians who have practiced in 60 countries, HealthBench includes 5,000 realistic health conversations, each with a custom physician-created rubric to grade model responses.
Task Categories:
Knowledge,Medical,QAEvaluation Metrics:
accuracy,communication_quality,completeness,context_awareness,instruction_followingAggregation Methods:
clipped_meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
communication,complex_responses,context_seeking,emergency_referrals,global_health,health_data_tasks,hedgingExtra Parameters:
{
"version": {
"type": "str",
"description": "Dataset file version, choices: ['Consensus', 'Hard', 'All'].",
"value": "Consensus",
"choices": [
"Consensus",
"Hard",
"All"
]
}
}
Prompt Template:
View
Answer the question:
{question}
HellaSwag#
Dataset Name:
hellaswagDataset ID: evalscope/hellaswag
Description:
HellaSwag is a benchmark for commonsense reasoning in natural language understanding tasks. It consists of multiple-choice questions where the model must select the most plausible continuation of a given context.
Task Categories:
Commonsense,Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
Humanity’s-Last-Exam#
Dataset Name:
hleDataset ID: cais/hle
Description:
Humanity’s Last Exam (HLE) is a language model benchmark consisting of 2,500 questions across a broad range of subjects. It was created jointly by the Center for AI Safety and Scale AI. The benchmark classifies the questions into the following broad subjects: mathematics (41%), physics (9%), biology/medicine (11%), humanities/social science (9%), computer science/artificial intelligence (10%), engineering (4%), chemistry (7%), and other (9%). Around 14% of the questions require the ability to understand both text and images, i.e., multi-modality. 24% of the questions are multiple-choice; the rest are short-answer, exact-match questions. To evaluate the performance of model without multi-modality capabilities, please set the
extra_params["include_multi_modal"]toFalse.Task Categories:
Knowledge,QAEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
Biology/Medicine,Chemistry,Computer Science/AI,Engineering,Humanities/Social Science,Math,Other,PhysicsExtra Parameters:
{
"include_multi_modal": {
"type": "bool",
"description": "Include multi-modal (image) questions during evaluation.",
"value": true
}
}
Prompt Template:
View
{question}
HumanEval#
Dataset Name:
humanevalDataset ID: opencompass/humaneval
Description:
HumanEval is a benchmark for evaluating the ability of code generation models to write Python functions based on given specifications. It consists of programming tasks with a defined input-output behavior. By default the code is executed in local environment. We recommend using sandbox execution to safely run and evaluate the generated code, please refer to the documentation for more details.
Task Categories:
CodingEvaluation Metrics:
Aggregation Methods:
mean_and_pass_at_kRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
openai_humanevalReview Timeout (seconds): 4
Sandbox Configuration:
{
"image": "python:3.11-slim",
"tools_config": {
"shell_executor": {},
"python_executor": {}
}
}
Prompt Template:
View
Read the following function signature and docstring, and fully implement the function described. Your response should only contain the code for this function.
{question}
IFBench#
Dataset Name:
ifbenchDataset ID: allenai/IFBench_test
Description:
IFBench is a new benchmark designed to evaluate how reliably AI models follow novel, challenging, and diverse verifiable instructions, with a strong focus on out-of-domain generalization. It comprises 58 manually curated verifiable constraints across categories such as counting, formatting, and word usage, aiming to address overfitting and data contamination issues present in existing benchmarks. Developed by AllenAI, IFBench serves as a rigorous test for precise instruction-following capabilities.
Task Categories:
InstructionFollowingEvaluation Metrics:
inst_level_loose,inst_level_strict,prompt_level_loose,prompt_level_strictAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
default
IFEval#
Dataset Name:
ifevalDataset ID: opencompass/ifeval
Description:
IFEval is a benchmark for evaluating instruction-following language models, focusing on their ability to understand and respond to various prompts. It includes a diverse set of tasks and metrics to assess model performance comprehensively.
Task Categories:
InstructionFollowingEvaluation Metrics:
inst_level_loose,inst_level_strict,prompt_level_loose,prompt_level_strictAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
default
IQuiz#
Dataset Name:
iquizDataset ID: AI-ModelScope/IQuiz
Description:
IQuiz is a benchmark for evaluating AI models on IQ and EQ questions. It consists of multiple-choice questions where the model must select the correct answer and provide an explanation.
Task Categories:
Chinese,Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
EQ,IQPrompt Template:
View
回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：[LETTER]"（不带引号），其中 [LETTER] 是 {letters} 中的一个。请在回答前进行一步步思考。
问题：{question}
选项：
{choices}
Live-Code-Bench#
Dataset Name:
live_code_benchDataset ID: AI-ModelScope/code_generation_lite
Description:
Live Code Bench is a benchmark for evaluating code generation models on real-world coding tasks. It includes a variety of programming problems with test cases to assess the model’s ability to generate correct and efficient code solutions. By default the code is executed in local environment. We recommend using sandbox execution to safely run and evaluate the generated code, please refer to the documentation for more details.
Task Categories:
CodingEvaluation Metrics:
Aggregation Methods:
mean_and_pass_at_kRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
release_latestReview Timeout (seconds): 6
Extra Parameters:
{
"start_date": {
"type": "str | null",
"description": "Filter problems starting from this date (YYYY-MM-DD). Null keeps all.",
"value": null
},
"end_date": {
"type": "str | null",
"description": "Filter problems up to this date (YYYY-MM-DD). Null keeps all.",
"value": null
},
"debug": {
"type": "bool",
"description": "Enable verbose debug logging and bypass certain safety checks.",
"value": false
}
}
Sandbox Configuration:
{
"image": "python:3.11-slim",
"tools_config": {
"shell_executor": {},
"python_executor": {}
}
}
Prompt Template:
View
### Question:
{question_content}
{format_prompt} ### Answer: (use the provided format with backticks)
LogiQA#
Dataset Name:
logi_qaDataset ID: extraordinarylab/logiqa
Description:
LogiQA is a dataset sourced from expert-written questions for testing human Logical reasoning.
Task Categories:
MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
MaritimeBench#
Dataset Name:
maritime_benchDataset ID: HiDolphin/MaritimeBench
Description:
MaritimeBench is a benchmark for evaluating AI models on maritime-related multiple-choice questions. It consists of questions related to maritime knowledge, where the model must select the correct answer from given options.
Task Categories:
Chinese,Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
请回答单选题。要求只输出选项，不输出解释，将选项放在[]里，直接输出答案。示例：
题目：在船舶主推进动力装置中，传动轴系在运转中承受以下复杂的应力和负荷，但不包括______。
选项：
A. 电磁力
B. 压拉应力
C. 弯曲应力
D. 扭应力
答：[A]
当前题目
{question}
选项：
{choices}
MATH-500#
Dataset Name:
math_500Dataset ID: AI-ModelScope/MATH-500
Description:
MATH-500 is a benchmark for evaluating mathematical reasoning capabilities of AI models. It consists of 500 diverse math problems across five levels of difficulty, designed to test a model’s ability to solve complex mathematical problems by generating step-by-step solutions and providing the correct final answer.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
Level 1,Level 2,Level 3,Level 4,Level 5Prompt Template:
View
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
MathQA#
Dataset Name:
math_qaDataset ID: extraordinarylab/math-qa
Description:
MathQA dataset is gathered by using a new representation language to annotate over the AQuA-RAT dataset with fully-specified operational programs.
Task Categories:
MCQ,Math,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
Med-MCQA#
Dataset Name:
med_mcqaDataset ID: extraordinarylab/medmcqa
Description:
MedMCQA is a large-scale MCQA dataset designed to address real-world medical entrance exam questions.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
MGSM#
Dataset Name:
mgsmDataset ID: evalscope/mgsm
Description:
Multilingual Grade School Math Benchmark (MGSM) is a benchmark of grade-school math problems, proposed in the paper Language models are multilingual chain-of-thought reasoners.
Task Categories:
Math,MultiLingual,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 4-shot
Subsets:
bn,de,en,es,fr,ja,ru,sw,te,th,zhPrompt Template:
View
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
Minerva-Math#
Dataset Name:
minerva_mathDataset ID: knoveleng/Minerva-Math
Description:
Minerva-math is a benchmark designed to evaluate the mathematical and quantitative reasoning capabilities of LLMs. It consists of 272 problems sourced primarily from MIT OpenCourseWare courses, covering advanced STEM subjects such as solid-state chemistry, astronomy, differential equations, and special relativity at the university and graduate level.
Task Categories:
Math,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
{question}
Please reason step by step, and put your final answer within \boxed{{}}.
MIT-Movie-Trivia#
Dataset Name:
mit_movie_triviaDataset ID: extraordinarylab/mit-movie-trivia
Description:
The MIT-Movie-Trivia dataset, originally created for slot filling, is modified by ignoring some slot types (e.g. genre, rating) and merging others (e.g. director and actor in person, and song and movie title in title) in order to keep consistent named entity types across all datasets.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
MIT-Restaurant#
Dataset Name:
mit_restaurantDataset ID: extraordinarylab/mit-restaurant
Description:
The MIT-Restaurant dataset is a collection of restaurant review text specifically curated for training and testing Natural Language Processing (NLP) models, particularly for Named Entity Recognition (NER). It contains sentences from real reviews, along with corresponding labels in the BIO format.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
MMLU#
Dataset Name:
mmluDataset ID: cais/mmlu
Description:
The MMLU (Massive Multitask Language Understanding) benchmark is a comprehensive evaluation suite designed to assess the performance of language models across a wide range of subjects and tasks. It includes multiple-choice questions from various domains, such as history, science, mathematics, and more, providing a robust measure of a model’s understanding and reasoning capabilities.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
abstract_algebra,anatomy,astronomy,business_ethics,clinical_knowledge,college_biology,college_chemistry,college_computer_science,college_mathematics,college_medicine,college_physics,computer_security,conceptual_physics,econometrics,electrical_engineering,elementary_mathematics,formal_logic,global_facts,high_school_biology,high_school_chemistry,high_school_computer_science,high_school_european_history,high_school_geography,high_school_government_and_politics,high_school_macroeconomics,high_school_mathematics,high_school_microeconomics,high_school_physics,high_school_psychology,high_school_statistics,high_school_us_history,high_school_world_history,human_aging,human_sexuality,international_law,jurisprudence,logical_fallacies,machine_learning,management,marketing,medical_genetics,miscellaneous,moral_disputes,moral_scenarios,nutrition,philosophy,prehistory,professional_accounting,professional_law,professional_medicine,professional_psychology,public_relations,security_studies,sociology,us_foreign_policy,virology,world_religionsPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
MMLU-Pro#
Dataset Name:
mmlu_proDataset ID: TIGER-Lab/MMLU-Pro
Description:
MMLU-Pro is a benchmark for evaluating language models on multiple-choice questions across various subjects. It includes questions from different domains, where the model must select the correct answer from given options.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
biology,business,chemistry,computer science,economics,engineering,health,history,law,math,other,philosophy,physics,psychologyPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
Question:
{question}
Options:
{choices}
MMLU-Redux#
Dataset Name:
mmlu_reduxDataset ID: AI-ModelScope/mmlu-redux-2.0
Description:
MMLU-Redux is a benchmark for evaluating language models on multiple-choice questions across various subjects. It includes questions from different domains, where the model must select the correct answer from given options. The bad answers are corrected.
Task Categories:
Knowledge,MCQEvaluation Metrics:
{'acc': {'allow_inclusion': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
abstract_algebra,anatomy,astronomy,business_ethics,clinical_knowledge,college_biology,college_chemistry,college_computer_science,college_mathematics,college_medicine,college_physics,computer_security,conceptual_physics,econometrics,electrical_engineering,elementary_mathematics,formal_logic,global_facts,high_school_biology,high_school_chemistry,high_school_computer_science,high_school_european_history,high_school_geography,high_school_government_and_politics,high_school_macroeconomics,high_school_mathematics,high_school_microeconomics,high_school_physics,high_school_psychology,high_school_statistics,high_school_us_history,high_school_world_history,human_aging,human_sexuality,international_law,jurisprudence,logical_fallacies,machine_learning,management,marketing,medical_genetics,miscellaneous,moral_disputes,moral_scenarios,nutrition,philosophy,prehistory,professional_accounting,professional_law,professional_medicine,professional_psychology,public_relations,security_studies,sociology,us_foreign_policy,virology,world_religionsPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
MRI-MCQA#
Dataset Name:
mri_mcqaDataset ID: extraordinarylab/mri-mcqa
Description:
MRI-MCQA is a benchmark composed by multiple-choice questions related to Magnetic Resonance Imaging (MRI).
Task Categories:
Knowledge,MCQ,MedicalEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
Multi-IF#
Dataset Name:
multi_ifDataset ID: facebook/Multi-IF
Description:
Multi-IF is a benchmark designed to evaluate the performance of LLM models’ capabilities in multi-turn instruction following within a multilingual environment.
Task Categories:
InstructionFollowing,MultiLingual,MultiTurnEvaluation Metrics:
inst_level_loose,inst_level_strict,prompt_level_loose,prompt_level_strictAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
Chinese,English,French,German,Hindi,Italian,Portuguese,Russian,Spanish,Thai,VietnameseExtra Parameters:
{
"max_turns": {
"type": "int",
"description": "Maximum number of interactive turns to evaluate (1-3).",
"value": 3,
"choices": [
1,
2,
3
]
}
}
MusicTrivia#
Dataset Name:
music_triviaDataset ID: extraordinarylab/music-trivia
Description:
MusicTrivia is a curated dataset of multiple-choice questions covering both classical and modern music topics. It includes questions about composers, musical periods, and popular artists, designed for evaluating factual recall and domain-specific music knowledge.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
MuSR#
Dataset Name:
musrDataset ID: AI-ModelScope/MuSR
Description:
MuSR is a benchmark for evaluating AI models on multiple-choice questions related to murder mysteries, object placements, and team allocation.
Task Categories:
MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
murder_mysteries,object_placements,team_allocationPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
Needle-in-a-Haystack#
Dataset Name:
needle_haystackDataset ID: AI-ModelScope/Needle-in-a-Haystack-Corpus
Description:
Needle in a Haystack is a benchmark focused on information retrieval tasks. It requires the model to find specific information within a large corpus of text. Usage Example
Task Categories:
LongContext,RetrievalEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
chinese,englishExtra Parameters:
{
"retrieval_question": {
"type": "str",
"description": "Question used for retrieval evaluation.",
"value": "What is the best thing to do in San Francisco?"
},
"needles": {
"type": "list[str]",
"description": "List of factual needle strings inserted into the context.",
"value": [
"\nThe best thing to do in San Francisco is eat a sandwich and sit in Dolores Park on a sunny day.\n"
]
},
"context_lengths_min": {
"type": "int",
"description": "Minimum context length (tokens) to generate synthetic samples.",
"value": 1000
},
"context_lengths_max": {
"type": "int",
"description": "Maximum context length (tokens) to generate synthetic samples.",
"value": 32000
},
"context_lengths_num_intervals": {
"type": "int",
"description": "Number of intervals between min and max context lengths.",
"value": 10
},
"document_depth_percent_min": {
"type": "int",
"description": "Minimum insertion depth percentage for needles.",
"value": 0
},
"document_depth_percent_max": {
"type": "int",
"description": "Maximum insertion depth percentage for needles.",
"value": 100
},
"document_depth_percent_intervals": {
"type": "int",
"description": "Number of intervals between min and max depth percentages.",
"value": 10
},
"tokenizer_path": {
"type": "str",
"description": "Tokenizer checkpoint path used for tokenization.",
"value": "Qwen/Qwen3-0.6B"
},
"show_score": {
"type": "bool",
"description": "Render numerical scores on heatmap output images.",
"value": false
}
}
System Prompt:
View
You are a helpful AI bot that answers questions for a user. Keep your response short and direct
View
Please read the following text and answer the question below.
<text>
{context}
</text>
<question>
{question}
</question>
Don't give information outside the document or repeat your findings.
OntoNotes5#
Dataset Name:
ontonotes5Dataset ID: extraordinarylab/ontonotes5
Description:
OntoNotes Release 5.0 is a large, multilingual corpus containing text in English, Chinese, and Arabic across various genres like news, weblogs, and broadcast conversations. It is richly annotated with multiple layers of linguistic information, including syntax, predicate-argument structure, word sense, named entities, and coreference to support research and development in natural language processing.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}
OpenAI MRCR#
Dataset Name:
openai_mrcrDataset ID: openai-mirror/mrcr
Description:
Memory-Recall with Contextual Retrieval (MRCR). Evaluates retrieval and recall in long contexts by placing 2, 4 or 8 needles in the prompt. Measures whether the model can correctly extract and use them.
Task Categories:
LongContext,RetrievalEvaluation Metrics:
mrcr_scoreAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultExtra Parameters:
{
"max_context_size": {
"type": "int | null",
"description": "Maximum context tokens; samples exceeding are skipped. Defaults to None (no limit).",
"value": null
},
"needle_count": {
"type": "list[int] | null",
"description": "Needle count filter (allowed: 2,4,8). Must be a list, e.g., [2], [4], or [2, 4, 8]. None keeps all.",
"value": null
},
"tik_enc": {
"type": "str",
"description": "tiktoken encoding name used for token counting.",
"value": "o200k_base"
}
}
PIQA#
Dataset Name:
piqaDataset ID: extraordinarylab/piqa
Description:
PIQA addresses the challenging task of reasoning about physical commonsense in natural language.
Task Categories:
Commonsense,MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
PolyMath#
Dataset Name:
poly_mathDataset ID: evalscope/PolyMath
Description:
PolyMath is a multilingual mathematical reasoning benchmark covering 18 languages and 4 easy-to-hard difficulty levels, with 9,000 high-quality problem samples. Our benchmark ensures difficulty comprehensiveness, language diversity, and high-quality translation, making it a highly discriminative multilingual mathematical benchmark in the era of reasoning LLMs.
Task Categories:
Math,MultiLingual,ReasoningEvaluation Metrics:
{'acc': {'numeric': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
ar,bn,de,en,es,fr,id,it,ja,ko,ms,pt,ru,sw,te,th,vi,zhPrompt Template:
View
{question}
ProcessBench#
Dataset Name:
process_benchDataset ID: Qwen/ProcessBench
Description:
ProcessBench is a benchmark for evaluating AI models on mathematical reasoning tasks. It includes various subsets such as GSM8K, Math, OlympiadBench, and OmniMath, each with its own set of problems that require step-by-step reasoning to arrive at the correct answer.
Task Categories:
Math,ReasoningEvaluation Metrics:
correct_acc,error_acc,simple_f1_scoreAggregation Methods:
f1Requires LLM Judge: No
Default Shots: 0-shot
Subsets:
gsm8k,math,olympiadbench,omnimathPrompt Template:
View
CThe following is a math problem and a solution (split into paragraphs, enclosed with tags and indexed from 0):
[Math Problem]
{problem}
[Solution]
{tagged_response}
Your task is to review and critique the solution paragraph by paragraph. Once you identify an error in a paragraph, return the index of the paragraph where the earliest error occurs. Otherwise, return the index of -1 (which typically denotes "not found").
Please put your final answer (i.e., the index) in oxed{{}}.
PubMedQA#
Dataset Name:
pubmedqaDataset ID: extraordinarylab/pubmed-qa
Description:
PubMedQA reasons over biomedical research texts to answer the multiple-choice questions.
Task Categories:
Knowledge,Yes/NoEvaluation Metrics:
accuracy,f1_score,maybe_ratio,precision,recall,yes_ratioAggregation Methods:
f1Requires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
{question}
Please answer YES or NO or MAYBE without an explanation.
QASC#
Dataset Name:
qascDataset ID: extraordinarylab/qasc
Description:
QASC is a question-answering dataset with a focus on sentence composition. It consists of 9,980 8-way multiple-choice questions about grade school science.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
RACE#
Dataset Name:
raceDataset ID: evalscope/race
Description:
RACE is a benchmark for testing reading comprehension and reasoning abilities of neural models. It is constructed from Chinese middle and high school examinations.
Task Categories:
MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 3-shot
Subsets:
high,middlePrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
SciCode#
Dataset Name:
scicodeDataset ID: evalscope/SciCode
Description:
SciCode is a challenging benchmark designed to evaluate the capabilities of language models (LMs) in generating code for solving realistic scientific research problems. It has a diverse coverage of 16 subdomains from 5 domains: Physics, Math, Material Science, Biology, and Chemistry. Unlike previous benchmarks that consist of exam-like question-answer pairs, SciCode is converted from real research problems. SciCode problems naturally factorize into multiple subproblems, each involving knowledge recall, reasoning, and code synthesis. Sandbox environment is needed for execution to safely run and evaluate the generated code, please refer to the documentation for more details.
Task Categories:
CodingEvaluation Metrics:
main_problem_pass_rate,subproblem_pass_rateAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultReview Timeout (seconds): 300
Extra Parameters:
{
"provide_background": {
"type": "bool",
"value": false,
"description": "Include scientific background information written by scientists for the problem in the model's prompt."
}
}
Sandbox Configuration:
{
"image": "scicode-benchmark:latest",
"tools_config": {
"shell_executor": {},
"python_executor": {}
}
}
System Prompt:
View
PROBLEM DESCRIPTION:
You will be provided with a description of a scientific problem. You will solve these problems by solving a sequence of *subproblems*. The solution to each subproblem may be implemented using your solutions to earlier subproblems. Each subproblem should be solved by providing a Python function that meets the specifications provided.
For each subproblem, you will be provided with the following
1. a description of the subproblem
2. a function header, which you must use in your solution implementation
3. a return line, which you must use in your solution implementation
You must only use the following dependencies to implement your solution:
{required_dependencies}
You MUST NOT import these dependencies anywhere in the code you generate.
For each subproblem provided you must solve it as follows:
1. Generate scientific background required for the next step, in a comment
2. Implement a function to solve the problem provided, using the provided header and return line
The response must be formatted as ```python```
View
Implement code to solve the following subproblem, using the description, function header, and return line provided.
Remember that you may use functions that you generated previously as solutions to previous subproblems to implement your answer.
Remember that you MUST NOT include code to import dependencies.
Remember to ensure your response is in the format of ```python``` and includes necessary background as a comment at the top.
SUBPROBLEM DESCRIPTION:
{step_description_prompt}
FUNCTION HEADER:
{function_header}
RETURN LINE:
{return_line}
Example:
```python
# Background: [Here, insert the necessary scientific knowledge required for the next step.]
[Insert the Python code here based on the provided function header and dependencies.]
```
SciQ#
Dataset Name:
sciqDataset ID: extraordinarylab/sciq
Description:
The SciQ dataset contains crowdsourced science exam questions about Physics, Chemistry and Biology, among others. For the majority of the questions, an additional paragraph with supporting evidence for the correct answer is provided.
Task Categories:
Knowledge,MCQ,ReadingComprehensionEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
SimpleQA#
Dataset Name:
simple_qaDataset ID: evalscope/SimpleQA
Description:
SimpleQA is a benchmark designed to evaluate the performance of language models on simple question-answering tasks. It includes a set of straightforward questions that require basic reasoning and understanding capabilities.
Task Categories:
Knowledge,QAEvaluation Metrics:
is_correct,is_incorrect,is_not_attemptedAggregation Methods:
meanRequires LLM Judge: Yes
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the question:
{question}
SIQA#
Dataset Name:
siqaDataset ID: extraordinarylab/siqa
Description:
Social Interaction QA (SIQA) is a question-answering benchmark for testing social commonsense intelligence. Contrary to many prior benchmarks that focus on physical or taxonomic knowledge, Social IQa focuses on reasoning about people’s actions and their social implications.
Task Categories:
Commonsense,MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
SuperGPQA#
Dataset Name:
super_gpqaDataset ID: m-a-p/SuperGPQA
Description:
SuperGPQA is a large-scale multiple-choice question answering dataset, designed to evaluate the generalization ability of models across different fields. It contains 100,000+ questions from 50+ fields, with each question having 10 options.
Task Categories:
Knowledge,MCQEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
Aeronautical and Astronautical Science and Technology,Agricultural Engineering,Animal Husbandry,Applied Economics,Aquaculture,Architecture,Art Studies,Astronomy,Atmospheric Science,Basic Medicine,Biology,Business Administration,Chemical Engineering and Technology,Chemistry,Civil Engineering,Clinical Medicine,Computer Science and Technology,Control Science and Engineering,Crop Science,Education,Electrical Engineering,Electronic Science and Technology,Environmental Science and Engineering,Food Science and Engineering,Forestry Engineering,Forestry,Geography,Geological Resources and Geological Engineering,Geology,Geophysics,History,Hydraulic Engineering,Information and Communication Engineering,Instrument Science and Technology,Journalism and Communication,Language and Literature,Law,Library, Information and Archival Management,Management Science and Engineering,Materials Science and Engineering,Mathematics,Mechanical Engineering,Mechanics,Metallurgical Engineering,Military Science,Mining Engineering,Musicology,Naval Architecture and Ocean Engineering,Nuclear Science and Technology,Oceanography,Optical Engineering,Petroleum and Natural Gas Engineering,Pharmacy,Philosophy,Physical Education,Physical Oceanography,Physics,Political Science,Power Engineering and Engineering Thermophysics,Psychology,Public Administration,Public Health and Preventive Medicine,Sociology,Stomatology,Surveying and Mapping Science and Technology,Systems Science,Textile Science and Engineering,Theoretical Economics,Traditional Chinese Medicine,Transportation Engineering,Veterinary Medicine,Weapon Science and TechnologyPrompt Template:
View
Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.
{question}
{choices}
SWE-bench_Lite#
Dataset Name:
swe_bench_liteDataset ID: princeton-nlp/SWE-bench_Lite
Description:
SWE-bench Lite is subset of SWE-bench, a dataset that tests systems’ ability to solve GitHub issues automatically. The dataset collects 300 test Issue-Pull Request pairs from 11 popular Python. Evaluation is performed by unit test verification using post-PR behavior as the reference solution. Need to run
pip install swebench==4.1.0before evaluating. Usage ExampleTask Categories:
CodingEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultExtra Parameters:
{
"build_docker_images": {
"type": "bool",
"description": "Build Docker images locally for each sample.",
"value": true
},
"pull_remote_images_if_available": {
"type": "bool",
"description": "Attempt to pull existing remote Docker images before building.",
"value": true
},
"inference_dataset_id": {
"type": "str",
"description": "Oracle dataset ID used to fetch inference context.",
"value": "princeton-nlp/SWE-bench_oracle"
}
}
Prompt Template:
View
{question}
SWE-bench_Verified#
Dataset Name:
swe_bench_verifiedDataset ID: princeton-nlp/SWE-bench_Verified
Description:
SWE-bench Verified is a subset of 500 samples from the SWE-bench test set, which have been human-validated for quality. SWE-bench is a dataset that tests systems’ ability to solve GitHub issues automatically. Need to run
pip install swebench==4.1.0before evaluating. Usage ExampleTask Categories:
CodingEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultExtra Parameters:
{
"inference_dataset_id": {
"type": "str",
"description": "Oracle dataset ID used to fetch inference context.",
"value": "princeton-nlp/SWE-bench_oracle"
},
"build_docker_images": {
"type": "bool",
"description": "Build Docker images locally for each sample.",
"value": true
},
"pull_remote_images_if_available": {
"type": "bool",
"description": "Attempt to pull existing remote Docker images before building.",
"value": true
}
}
Prompt Template:
View
{question}
SWE-bench_Verified_mini#
Dataset Name:
swe_bench_verified_miniDataset ID: evalscope/swe-bench-verified-mini
Description:
SWEBench-verified-mini is a subset of SWEBench-verified that uses 50 instead of 500 datapoints, requires 5GB instead of 130GB of storage and has approximately the same distribution of performance, test pass rates and difficulty as the original dataset. Need to run
pip install swebench==4.1.0before evaluating. Usage ExampleTask Categories:
CodingEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultExtra Parameters:
{
"build_docker_images": {
"type": "bool",
"description": "Build Docker images locally for each sample.",
"value": true
},
"pull_remote_images_if_available": {
"type": "bool",
"description": "Attempt to pull existing remote Docker images before building.",
"value": true
},
"inference_dataset_id": {
"type": "str",
"description": "Oracle dataset ID used to fetch inference context.",
"value": "princeton-nlp/SWE-bench_oracle"
}
}
Prompt Template:
View
{question}
TriviaQA#
Dataset Name:
trivia_qaDataset ID: evalscope/trivia_qa
Description:
TriviaQA is a large-scale reading comprehension dataset consisting of question-answer pairs collected from trivia websites. It includes questions with multiple possible answers, making it suitable for evaluating the ability of models to understand and generate answers based on context.
Task Categories:
QA,ReadingComprehensionEvaluation Metrics:
{'acc': {'allow_inclusion': True}}Aggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
rc.wikipediaPrompt Template:
View
Read the content and answer the following question.
Content: {content}
Question: {question}
Keep your The last line of your response should be of the form "ANSWER: [ANSWER]" (without quotes) where [ANSWER] is the answer to the problem.
TruthfulQA#
Dataset Name:
truthful_qaDataset ID: evalscope/truthful_qa
Description:
TruthfulQA is a benchmark designed to evaluate the ability of AI models to answer questions truthfully and accurately. It includes multiple-choice tasks, focusing on the model’s understanding of factual information.
Task Categories:
KnowledgeEvaluation Metrics:
multi_choice_accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
multiple_choiceExtra Parameters:
{
"multiple_correct": {
"type": "bool",
"description": "Use multiple-answer format (MC2) if True; otherwise single-answer (MC1).",
"value": false
}
}
Prompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
Winogrande#
Dataset Name:
winograndeDataset ID: AI-ModelScope/winogrande_val
Description:
Winogrande is a benchmark for evaluating AI models on commonsense reasoning tasks, specifically designed to test the ability to resolve ambiguous pronouns in sentences.
Task Categories:
MCQ,ReasoningEvaluation Metrics:
accAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
defaultPrompt Template:
View
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
WMT2024++#
Dataset Name:
wmt24ppDataset ID: extraordinarylab/wmt24pp
Description:
WMT2024 news translation benchmark supporting multiple language pairs. Each subset represents a specific translation direction
Task Categories:
MachineTranslation,MultiLingualEvaluation Metrics:
bert_score,bleu,cometAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 0-shot
Subsets:
en-ar_eg,en-ar_sa,en-bg_bg,en-bn_in,en-ca_es,en-cs_cz,en-da_dk,en-de_de,en-el_gr,en-es_mx,en-et_ee,en-fa_ir,en-fi_fi,en-fil_ph,en-fr_ca,en-fr_fr,en-gu_in,en-he_il,en-hi_in,en-hr_hr,en-hu_hu,en-id_id,en-is_is,en-it_it,en-ja_jp,en-kn_in,en-ko_kr,en-lt_lt,en-lv_lv,en-ml_in,en-mr_in,en-nl_nl,en-no_no,en-pa_in,en-pl_pl,en-pt_br,en-pt_pt,en-ro_ro,en-ru_ru,en-sk_sk,en-sl_si,en-sr_rs,en-sv_se,en-sw_ke,en-sw_tz,en-ta_in,en-te_in,en-th_th,en-tr_tr,en-uk_ua,en-ur_pk,en-vi_vn,en-zh_cn,en-zh_tw,en-zu_zaPrompt Template:
View
Translate the following {source_language} sentence into {target_language}:
{source_language}: {source_text}
{target_language}:
WNUT2017#
Dataset Name:
wnut2017Dataset ID: extraordinarylab/wnut2017
Description:
The WNUT2017 dataset is a collection of user-generated text from various social media platforms, like Twitter and YouTube, specifically designed for a named-entity recognition task.
Task Categories:
Knowledge,NEREvaluation Metrics:
accuracy,f1_score,precision,recallAggregation Methods:
meanRequires LLM Judge: No
Default Shots: 5-shot
Subsets:
defaultPrompt Template:
View
You are a named entity recognition system that identifies the following entity types:
{entities}
Process the provided text and mark all named entities with XML-style tags.
For example:
<person>John Smith</person> works at <organization>Google</organization> in <location>Mountain View</location>.
Available entity tags: {entity_list}
INSTRUCTIONS:
1. Wrap your entire response in <response>...</response> tags.
2. Inside these tags, include the original text with entity tags inserted.
3. Do not change the original text in any way (preserve spacing, punctuation, case, etc.).
4. Tag ALL entities you can identify using the exact tag names provided.
5. Do not include explanations, just the tagged text.
6. If entity spans overlap, choose the most specific entity type.
7. Ensure every opening tag has a matching closing tag.
Text to process:
{text}