File size: 1,609 Bytes
da804e0 19e7282 42c2d33 f91d54c 42c2d33 73353ee 42c2d33 f91d54c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 |
A natural way to evaluate code programs is to see if they pass unit tests, it is the idea behind the [pass@k](https://huggingface.co/metrics/code_eval) metric, a popular evaluation framework for code generation models, on [HumanEval](https://huggingface.co/datasets/openai_humaneval) dataset, which was introduced in [Codex paper](https://arxiv.org/pdf/2107.03374v2.pdf). The dataset includes 164 handwritten programming problems. In the pass@k metric, k code samples are generated per problem, and a problem is considered solved if any sample passes the unit tests and the total fraction of problems solved is reported.
In most papers, 200 candidate program completions are sampled, and pass@1, pass@10, and pass@100 are computed using an unbiased sampling estimator. Table 1 below shows the HumanEval scores of CodeParrot, InCoder, PolyCoder, CodeGen and Codex (not open-source).
<div align="center">
Model | pass@1 | pass@10 | pass@100|
|-------|--------|---------|---------|
|CodeParrot (110M) | 3.80% | 6.57% | 12.78% |
|CodeParrot (1.5B) | 3.58% | 8.03% | 14.96% |
|||||
|InCoder (6.7B) | 15.2% | 27.8% | 47.00% |
|||||
|PolyCoder (160M)| 2.13% | 3.35% | 4.88% |
|PolyCoder (400M)| 2.96% | 5.29% | 11.59% |
|PolyCoder (2.7B)| 5.59% | 9.84% | 17.68% |
|||||
|CodeGen-Mono (350M)| 12.76% | 23.11% | 35.19% |
|CodeGen-Mono (2.7B)| 23.70% | 36.64% | 57.01% |
|CodeGen-Mono (6.1B)| 26.13% | 42.29% | 65.82% |
|CodeGen-Mono (16.1B)| **29.28%** | **49.86%** | **75.00%** |
|||||
|Codex (25M)| 3.21% | 7.1% | 12.89%|
|Codex (300M)| 13.17%| 20.37% | 36.27% |
|Codex (12B)| 28.81%| 46.81% | 72.31% |
</div> |