🎯 DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

📝 Paper@arXiv | 🤗 Datasets&Models@HF | 🐱 Code@GitHub 🐦 Thread@X(Twitter) | 🐶 中文博客@知乎 | 📊 Leaderboard@PapersWithCode | 📑 BibTeX

[!IMPORTANT] 🔥 Excited to find our DART-Math-DSMath-7B (Prop2Diff) trained on DART-Math-Hard comparable to the AIMO winner NuminaMath-7B on CoT, but based solely on MATH & GSM8K prompt set, leaving much room to improve! Besides, our DART method is also fully compatible… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/dart-math-hard.

🎯 DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

📝 Paper@arXiv | 🤗 Datasets&Models@HF | 🐱 Code@GitHub 🐦 Thread@X(Twitter) | 🐶 中文博客@知乎 | 📊 Leaderboard@PapersWithCode | 📑 BibTeX

  Datasets: DART-Math

DART-Math datasets are the state-of-the-art and data-efficientopen-source instruction tuning datasets for mathematical reasoning.

Figure 1: Left: Average accuracy on 6 mathematical benchmarks. We compare with models… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/dart-math-uniform.

[!NOTE] This dataset is the data pool synthesized from the query set of the MATH training set, containing all answer-correct samples and other metadata produced during the work. DART-Math-* datasets are extracted from dart-math-pool-* data pools.

  🎯 DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

📝 Paper@arXiv | 🤗 Datasets&Models@HF | 🐱 Code@GitHub 🐦 Thread@X(Twitter) | 🐶 中文博客@知乎 | 📊 Leaderboard@PapersWithCode | 📑 BibTeX

  Datasets:… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/dart-math-pool-math.

Litzy619/dart-math-diff dataset hosted on Hugging Face and contributed by the HF Datasets community

[!NOTE] This dataset is the synthesis information of queries from the GSM8K training set, such as the numbers of raw/correct samples of each synthesis job. Usually used with dart-math-pool-gsm8k.

  🎯 DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

📝 Paper@arXiv | 🤗 Datasets&Models@HF | 🐱 Code@GitHub 🐦 Thread@X(Twitter) | 🐶 中文博客@知乎 | 📊 Leaderboard@PapersWithCode | 📑 BibTeX

  Datasets: DART-Math

DART-Math datasets are the… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/dart-math-pool-gsm8k-query-info.

Dataset Card for "dart-math-uniform"

More Information needed

[!NOTE] This dataset is the synthesis information of queries from the MATH training set, such as the numbers of raw/correct samples of each synthesis job. Usually used with dart-math-pool-math.

  🎯 DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

📝 Paper@arXiv | 🤗 Datasets&Models@HF | 🐱 Code@GitHub 🐦 Thread@X(Twitter) | 🐶 中文博客@知乎 | 📊 Leaderboard@PapersWithCode | 📑 BibTeX

  Datasets: DART-Math

DART-Math datasets are the state-of-the-art… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/dart-math-pool-math-query-info.

Dart-math-hard_scored - with OpenDataArena Scores

This dataset is a scored version of the original hkust-nlp/dart-math-hard dataset. The scoring was performed using the OpenDataArena-Tool, a comprehensive suite of automated evaluation methods for assessing instruction-following datasets. This version of the dataset includes rich, multi-dimensional scores for both the instructions (questions) and the instruction-response pairs, allowing for highly granular data analysis and… See the full description on the dataset page: https://huggingface.co/datasets/OpenDataArena/hkust-nlp_dart-math-hard_scored.

Dataset Card for "rlhflow_mixture_mod_scalebiosampled-20k"

weight = { 'MathInstruct': 0.17918314039707184, 'SlimOrca': 0.1572466790676117, 'Magicoder-Evol-Instruct-110K': 0.1262860894203186, 'dart-math-uniform': 0.10912656784057617, 'GPTeacher-General-Instruct': 0.10593341290950775, 'GPT4-LLM-Cleaned': 0.09206369519233704, 'WizardLM_evol_instruct_V2_196k': 0.07409033179283142, 'UltraInteract_sft': 0.056401610374450684, 'orca-math-word-problems-200k': 0.054032713174819946… See the full description on the dataset page: https://huggingface.co/datasets/pxyyy/rlhflow_mixture_mod_scalebiosampled-20k.

Dataset Card for "rlhflow_mixture_intuitive_sampled-20k"

weight = { 'MathInstruct': 0.1, 'SlimOrca': 0.2, 'Magicoder-Evol-Instruct-110K': 0.1, 'dart-math-uniform': 0.3, 'GPTeacher-General-Instruct': 0.05, 'GPT4-LLM-Cleaned': 0.03, 'WizardLM_evol_instruct_V2_196k': 0.05, 'UltraInteract_sft': 0.1, 'orca-math-word-problems-200k': 0.05, 'ShareGPT_V3_unfiltered_cleaned_split_no_imsorry': 0.02, }

More Information needed

GSM8K (Fixed)

Some erroneous labels exist in the GSM8K dataset. This dataset is fixed from https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/train.jsonl with the code appended at the end. The errors are located by delving into unreasonably low pass rates by the strong DeepSeekMath-7B-RL and hopefully should be exhaustive. This dataset is used by the 🎯DART-Math project to synthesize data.

[!WARNING] ⚠️ Only the training set has been fixed so far.

for… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/gsm8k-fix.

Dataset Card for "rlhflow_mixture_scalebio_sampled-nolisa-600k"

weight = { 'SlimOrca': 0.34525978565216064, 'dart-math-uniform': 0.23386941850185394, 'GPT4-LLM-Cleaned': 0.19111572206020355, 'MathInstruct': 0.16642746329307556, 'GPTeacher-General-Instruct': 0.042891550809144974, 'ShareGPT_V3_unfiltered_cleaned_split_no_imsorry': 0.006720397621393204, 'UltraInteract_sft': 0.0042861211113631725, 'WizardLM_evol_instruct_V2_196k': 0.004021201748400927, 'Magicoder-Evol-Instruct-110K':… See the full description on the dataset page: https://huggingface.co/datasets/pxyyy/rlhflow_mixture_scalebio_sampled-nolisa-250k.

[!NOTE] This dataset is the VRT baseline dataset used to train baseline models *-VRT in Table 2 of the paper.

Another ablation baseline to DART is vanilla rejection tuning (VRT), where we synthesize a dataset of the same size of 0.59M examples with DeepSeekMath-7B-RL, using vanilla rejection sampling as described in §2.1.

  🎯 DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

📝 Paper@arXiv | 🤗 Datasets&Models@HF | 🐱 Code@GitHub 🐦… See the full description on the dataset page: https://huggingface.co/datasets/hkust-nlp/vrt-baseline.