Note Performance produced by `DART-Math-Hard` is usually but not necessarily **slightly better (~1% absolutely)** than `DART-Math-Uniform`. `DART-Math-DSMath-7B` achieves 53.6% on MATH and 86.8% on GSM8K for in-domain benchmarks, and 40.7% on CollegeMath as a representative out-of-domain benchmark.

hkust-nlp/dart-math-dsmath-7b-prop2diff

Text Generation • Updated Jul 21 • 323 • 3

Note All the `DART-Math` models use the Alpaca prompt template: ``` Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n###Instruction:\n{query}\n\n### Response:\n ```

Note 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.

Note 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`.

Note Similar to `dart-math-pool-math`.

Note Similar to `dart-math-pool-math-query-info`.

Note The VRT baseline dataset used to train baseline models `*-VRT` in Table 2 of the paper.