Our goal with this group is to create an unchanging through time version of evaluations that will power the Open LLM Leaderboard on HuggingFace.
As we want to evaluate models across capabilities, the list currently contains:
- BBH (3-shots, multichoice)
- GPQA (0-shot, multichoice)
- mmlu-pro (5-shots, multichoice)
- Musr (0-shot, multichoice)
- ifeval (0-shot, generative)
- Math-lvl-5 (4-shots, generative, minerva version)
Details on the choice of those evals can be found here !
A suite of 23 challenging BIG-Bench tasks which we call BIG-Bench Hard (BBH). These are the task for which prior language model evaluations did not outperform the average human-rater.
Title: Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them
BIG-Bench (Srivastava et al., 2022) is a diverse evaluation suite that focuses on tasks believed to be beyond the capabilities of current language models. Language models have already made good progress on this benchmark, with the best model in the BIG-Bench paper outperforming average reported human-rater results on 65% of the BIG-Bench tasks via few-shot prompting. But on what tasks do language models fall short of average human-rater performance, and are those tasks actually unsolvable by current language models? In this work, we focus on a suite of 23 challenging BIG-Bench tasks which we call BIG-Bench Hard (BBH). These are the task for which prior language model evaluations did not outperform the average human-rater. We find that applying chain-of-thought (CoT) prompting to BBH tasks enables PaLM to surpass the average human-rater performance on 10 of the 23 tasks, and Codex (code-davinci-002) to surpass the average human-rater performance on 17 of the 23 tasks. Since many tasks in BBH require multi-step reasoning, few-shot prompting without CoT, as done in the BIG-Bench evaluations (Srivastava et al., 2022), substantially underestimates the best performance and capabilities of language models, which is better captured via CoT prompting. As further analysis, we explore the interaction between CoT and model scale on BBH, finding that CoT enables emergent task performance on several BBH tasks with otherwise flat scaling curves.
- paper: https://huggingface.co/papers/2210.09261
- Homepage: https://github.com/suzgunmirac/BIG-Bench-Hard
@article{suzgun2022challenging,
title={Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them},
author={Suzgun, Mirac and Scales, Nathan and Sch{\"a}rli, Nathanael and Gehrmann, Sebastian and Tay, Yi and Chung, Hyung Won and Chowdhery, Aakanksha and Le, Quoc V and Chi, Ed H and Zhou, Denny and and Wei, Jason},
journal={arXiv preprint arXiv:2210.09261},
year={2022}
}
leaderboard_bbh
leaderboard_bbh_boolean_expressions
leaderboard_bbh_causal_judgement
leaderboard_bbh_date_understanding
leaderboard_bbh_disambiguation_qa
leaderboard_bbh_formal_fallacies
leaderboard_bbh_geometric_shapes
leaderboard_bbh_hyperbaton
leaderboard_bbh_logical_deduction_five_objects
leaderboard_bbh_logical_deduction_seven_objects
leaderboard_bbh_logical_deduction_three_objects
leaderboard_bbh_movie_recommendation
leaderboard_bbh_navigate
leaderboard_bbh_object_counting
leaderboard_bbh_penguins_in_a_table
leaderboard_bbh_reasoning_about_colored_objects
leaderboard_bbh_ruin_names
leaderboard_bbh_salient_translation_error_detection
leaderboard_bbh_snarks
leaderboard_bbh_sports_understanding
leaderboard_bbh_temporal_sequences
leaderboard_bbh_tracking_shuffled_objects_five_objects
leaderboard_bbh_tracking_shuffled_objects_seven_objects
leaderboard_bbh_tracking_shuffled_objects_three_objects
leaderboard_bbh_web_of_lies
Title: GPQA: A Graduate-Level Google-Proof Q&A Benchmark
We present GPQA, a challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. We ensure that the questions are high-quality and extremely difficult: experts who have or are pursuing PhDs in the corresponding domains reach 65% accuracy (74% when discounting clear mistakes the experts identified in retrospect), while highly skilled non-expert validators only reach 34% accuracy, despite spending on average over 30 minutes with unrestricted access to the web (i.e., the questions are “Google-proof”). The questions are also difficult for state-of-the-art AI systems, with our strongest GPT-4–based baseline achieving 39% accuracy. If we are to use future AI systems to help us answer very hard questions—for example, when developing new scientific knowledge—we need to develop scalable oversight methods that enable humans to supervise their outputs, which may be difficult even if the supervisors are themselves skilled and knowledgeable. The difficulty of GPQA both for skilled non-experts and frontier AI systems should enable realistic scalable oversight experiments, which we hope can help devise ways for human experts to reliably get truthful information from AI systems that surpass human capabilities.
- Paper: https://huggingface.co/papers/2311.12022
- Homepage: https://github.com/idavidrein/gpqa/tree/main
@misc{rein2023gpqa,
title={GPQA: A Graduate-Level Google-Proof Q&A Benchmark},
author={David Rein and Betty Li Hou and Asa Cooper Stickland and Jackson Petty and Richard Yuanzhe Pang and Julien Dirani and Julian Michael and Samuel R. Bowman},
year={2023},
eprint={2311.12022},
archivePrefix={arXiv},
primaryClass={cs.AI}
}
leaderboard_gpqa
leaderboard_gpqa_extended
leaderboard_gpqa_diamond
leaderboard_gpqa_main
Title: Instruction-Following Evaluation for Large Language Models
One core capability of Large Language Models (LLMs) is to follow natural language instructions. However, the evaluation of such abilities is not standardized: Human evaluations are expensive, slow, and not objectively reproducible, while LLM-based auto-evaluation is potentially biased or limited by the ability of the evaluator LLM. To overcome these issues, we introduce Instruction-Following Eval (IFEval) for large language models. IFEval is a straightforward and easy-to-reproduce evaluation benchmark. It focuses on a set of "verifiable instructions" such as "write in more than 400 words" and "mention the keyword of AI at least 3 times". We identified 25 types of those verifiable instructions and constructed around 500 prompts, with each prompt containing one or more verifiable instructions. We show evaluation results of two widely available LLMs on the market.
- Paper: https://huggingface.co/papers/2210.09261
- Homepage: https://github.com/google-research/google-research/tree/master/instruction_following_eval
@article{zhou2023instructionfollowing,
title={Instruction-Following Evaluation for Large Language Models},
author={Jeffrey Zhou and Tianjian Lu and Swaroop Mishra and Siddhartha Brahma and Sujoy Basu and Yi Luan and Denny Zhou and Le Hou},
journal={arXiv preprint arXiv:2311.07911},
year={2023},
}
leaderboard_ifeval
This is the 4 shots variant of minerva math but only keeping the level 5 questions.
Title: Measuring Mathematical Problem Solving With the MATH Dataset
Many intellectual endeavors require mathematical problem solving, but this skill remains beyond the capabilities of computers. To measure this ability in machine learning models, we introduce MATH, a new dataset of 12,500 challenging competition mathematics problems. Each problem in MATH has a full step-by-step solution which can be used to teach models to generate answer derivations and explanations.
NOTE: The few-shot and the generated answer extraction is based on the
Minerva and exact match equivalence is
calculated using the sympy
library. This requires additional dependencies,
which can be installed via the lm-eval[math]
extra.
@article{hendrycksmath2021,
title={Measuring Mathematical Problem Solving With the MATH Dataset},
author={Dan Hendrycks and Collin Burns and Saurav Kadavath and Akul Arora and Steven Basart and Eric Tang and Dawn Song and Jacob Steinhardt},
journal={NeurIPS},
year={2021}
}
@misc{2206.14858,
Author = {Aitor Lewkowycz and Anders Andreassen and David Dohan and Ethan Dye and Henryk Michalewski and Vinay Ramasesh and Ambrose Slone and Cem Anil and Imanol Schlag and Theo Gutman-Solo and Yuhuai Wu and Behnam Neyshabur and Guy Gur-Ari and Vedant Misra},
Title = {Solving Quantitative Reasoning Problems with Language Models},
Year = {2022},
Eprint = {arXiv:2206.14858},
}
leaderboard_math_hard
leaderboard_math_algebra_hard
leaderboard_math_counting_and_prob_hard
leaderboard_math_geometry_hard
leaderboard_math_intermediate_algebra_hard
leaderboard_math_num_theory_hard
leaderboard_math_prealgebra_hard
leaderboard_math_precalculus_hard
Title: MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark
In the age of large-scale language models, benchmarks like the Massive Multitask Language Understanding (MMLU) have been pivotal in pushing the boundaries of what AI can achieve in language comprehension and reasoning across diverse domains. However, as models continue to improve, their performance on these benchmarks has begun to plateau, making it increasingly difficult to discern differences in model capabilities. This paper introduces MMLU-Pro, an enhanced dataset designed to extend the mostly knowledge-driven MMLU benchmark by integrating more challenging, reasoning-focused questions and expanding the choice set from four to ten options. Additionally, MMLU-Pro eliminates the trivial and noisy questions in MMLU. Our experimental results show that MMLU-Pro not only raises the challenge, causing a significant drop in accuracy by 16% to 33% compared to MMLU but also demonstrates greater stability under varying prompts. With 24 different prompt styles tested, the sensitivity of model scores to prompt variations decreased from 4-5% in MMLU to just 2% in MMLU-Pro. Additionally, we found that models utilizing Chain of Thought (CoT) reasoning achieved better performance on MMLU-Pro compared to direct answering, which is in stark contrast to the findings on the original MMLU, indicating that MMLU-Pro includes more complex reasoning questions. Our assessments confirm that MMLU-Pro is a more discriminative benchmark to better track progress in the field.
- Paper: https://huggingface.co/papers/2406.01574
- Homepage: https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro
@misc{wang2024mmluprorobustchallengingmultitask,
title={MMLU-Pro: A More Robust and Challenging Multi-Task Language
Understanding Benchmark},
author={Yubo Wang and Xueguang Ma and Ge Zhang and Yuansheng Ni and Abhranil Chandra and Shiguang Guo and Weiming Ren and Aaran Arulraj and Xuan He and Ziyan Jiang and Tianle Li and Max Ku and Kai Wang and Alex Zhuang and Rongqi Fan and Xiang Yue and Wenhu Chen},
year={2024},
eprint={2406.01574},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2406.01574},
}
leaderboard_mmlu_pro
leaderboard_mmlu_pro
Title: MuSR: Testing the Limits of Chain-of-thought with Multistep Soft Reasoning
While large language models (LLMs) equipped with techniques like chain-of-thought prompting have demonstrated impressive capabilities, they still fall short in their ability to reason robustly in complex settings. However, evaluating LLM reasoning is challenging because system capabilities continue to grow while benchmark datasets for tasks like logical deduction have remained static. We introduce MuSR, a dataset for evaluating language models on multistep soft reasoning tasks specified in a natural language narrative. This dataset has two crucial features. First, it is created through a novel neurosymbolic synthetic-to-natural generation algorithm, enabling the construction of complex reasoning instances that challenge GPT-4 (e.g., murder mysteries roughly 1000 words in length) and which can be scaled further as more capable LLMs are released. Second, our dataset instances are free text narratives corresponding to real-world domains of reasoning; this makes it simultaneously much more challenging than other synthetically-crafted benchmarks while remaining realistic and tractable for human annotators to solve with high accuracy. We evaluate a range of LLMs and prompting techniques on this dataset and characterize the gaps that remain for techniques like chain-of-thought to perform robust reasoning.
@misc{sprague2024musrtestinglimitschainofthought,
title={MuSR: Testing the Limits of Chain-of-thought with Multistep Soft
Reasoning},
author={Zayne Sprague and Xi Ye and Kaj Bostrom and Swarat Chaudhuri and Greg Durrett},
year={2024},
eprint={2310.16049},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2310.16049},
}
leaderboard_musr
leaderboard_musr_murder_mysteries
leaderboard_musr_object_placements
leaderboard_musr_team_allocation