What are the differences compared to the original Python version?

[findmyway]

• The prompt is more Julian. Each doc string is organized in the following format:

"""
    func_name(arg::ArgType, [...])::ReturnType

Problem description.

# Examples

```jldoctest
julia> func_name(...)
Expected Results
```
"""

Note

Note that I don't add an extra function signature line after the doc string in our prompt. The main reason is that by doing so most LLMs are tempted to generate results in the completion style instead of the chat style, leading to degraded performance.

• Type restrictions are added as much as possible based on the doc string.
• Fixed several docstring and test cases given that arrays in Julia are 1-based indexed by default.
• True/False -> true/false, ' -> ", BigInt, etc.
• Added extra test cases from evalplus
• Applied JuliaFormatter.jl by default.
• Example usages are translated to jldoctest style and are well tested.
• All the benchmark solutions are manually reimplemented based on the original Python version solution.