temporary0-0name commited on
Commit
61532b6
·
verified ·
1 Parent(s): 041def4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -4
README.md CHANGED
@@ -8,10 +8,12 @@ size_categories:
8
  ---
9
 
10
  <h1 style="text-align: center;">Abstract</h1>
11
- Driven by the surge in code generation using large language models (LLMs), numerous benchmarks have emerged to evaluate these LLMs capabilities. We conducted a large-scale human evaluation of HumanEval and MBPP, two popular benchmarks for Python code generation, analyzing their diversity and difficulty.
12
- Our findings unveil a critical bias towards a limited set of programming concepts, neglecting most of the other concepts entirely. Furthermore, we uncover a worrying prevalence of easy tasks that can inflate model performance estimations. To address these limitations, we propose a novel benchmark, PythonSaga, featuring 185 hand-crafted prompts in a balanced representation of 38 programming concepts across diverse difficulty levels.
13
- The robustness of our benchmark is demonstrated by the poor performance of existing Code-LLMs. The code and data set are openly available to the NLP community at [here](https://github.com/ PythonSaga/PythonSaga) .
14
- <br>
 
 
15
  <br>
16
 
17
  <h1 style="text-align: center;">PythonSaga</h1>
 
8
  ---
9
 
10
  <h1 style="text-align: center;">Abstract</h1>
11
+ <p>
12
+ Driven by the surge in code generation using large language models (LLMs), numerous benchmarks have emerged to evaluate these LLMs' capabilities. We conducted a large-scale human evaluation of HumanEval and MBPP, two popular benchmarks for Python code generation, analyzing their diversity and difficulty.
13
+ Our findings unveil a critical bias towards a limited set of programming concepts, neglecting most of the other concepts entirely. Furthermore, we uncover a worrying prevalence of easy tasks that can inflate model performance estimations. To address these limitations, we propose a novel benchmark, PythonSaga, featuring 185 hand-crafted prompts in a balanced representation of 38 programming concepts across diverse difficulty levels.
14
+ The robustness of our benchmark is demonstrated by the poor performance of existing Code-LLMs. The code and dataset are openly available to the NLP community at
15
+ <a href="https://github.com/PythonSaga/PythonSaga" target="_blank">https://github.com/PythonSaga/PythonSaga</a>.
16
+ </p>
17
  <br>
18
 
19
  <h1 style="text-align: center;">PythonSaga</h1>