GSM8K (Grade School Math 8K) Benchmark

A GSM8K (Grade School Math 8K) Benchmark is a mathematical reasoning benchmark with linguistically diverse grade school math word problems.

• Context:
  • It can (typically) contain 8,500 math word problems, divided into 7,500 training problems and 1,000 test problems, designed to test multi-step problem-solving abilities.
  • It can (often) involve basic arithmetic operations such as Addition, Subtraction, Multiplication, and Division, requiring between 2 to 8 steps to solve.
  • ...
  • It can serve as a benchmark to assess the mathematical reasoning of LLMs.
  • It can support research in improving AI multi-step reasoning, especially for natural language mathematical problems.
  • It can implement techniques like Chain-of-Thought Prompting to assist models in generating intermediate steps for better problem-solving.
  • It can challenge AI systems, as even the latest transformer models struggle to achieve high accuracy, highlighting limitations in current model architectures.
  • ...
• Example(s):
  • ...
• Counter-Example(s):
  • GSM-Symbolic.
  • Simple Arithmetic Datasets, which do not require multi-step reasoning or linguistic processing.
• See: Chain-of-Thought Prompting, Tree-of-Thought Prompting, Transformer Models, Benchmark Datasets

---

---

References

2024

• (Mirzadeh et al., 2024) ⇒ Iman Mirzadeh, Keivan Alizadeh, Hooman Shahrokhi, Oncel Tuzel, Samy Bengio, and Mehrdad Farajtabar. (2024). “GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models.” In: arXiv preprint arXiv:2410.05229.
  • NOTES
    • The paper introduces GSM-Symbolic, a new benchmark for evaluating mathematical reasoning in LLMs, addressing limitations of the GSM8K benchmark.

---