| from datasets import DatasetDict, load_dataset |
| from typing import Dict |
| |
| from data.base_builder import BaseBuilder |
| from data.kodcode.env import KodCodeEnv |
|
|
| class KodCodeBuilder(BaseBuilder): |
| |
| def get_env_cls(self): |
| return KodCodeEnv |
| |
| def _build_datasets(self) -> DatasetDict: |
| |
| all_dataset = load_dataset("KodCode/KodCode-Light-RL-10K") |
| all_correct_dataset = all_dataset["train"] |
| |
| |
| train_ratio, valid_ratio, test_ratio = self.config.get("train_ratio"), self.config.get("valid_ratio"), self.config.get("test_ratio") |
| assert train_ratio + valid_ratio + test_ratio == 1 |
| |
| all_size = len(all_correct_dataset) |
| test_size = int(all_size * test_ratio) |
| split = all_correct_dataset.train_test_split(test_size=test_size, shuffle=True) |
| train_valid_dataset, test_dataset = split["train"], split["test"] |
| |
| valid_size = int(len(train_valid_dataset) * valid_ratio / (train_ratio + valid_ratio)) |
| split = train_valid_dataset.train_test_split(test_size=valid_size, shuffle=True) |
| train_dataset, valid_dataset = split["train"], split["test"] |
| |
| |
| train_dataset = train_dataset.map(self._sft_preprocess).select_columns(self._sft_keep_keys()) |
| valid_dataset = valid_dataset.map(self._sft_preprocess).select_columns(self._sft_keep_keys()) |
| test_dataset = test_dataset.map(self._sft_preprocess).select_columns(self._sft_keep_keys()) |
| |
| |
| dataset_dict = DatasetDict() |
| dataset_dict["train"] = train_dataset |
| dataset_dict["valid"] = valid_dataset |
| dataset_dict["test"] = test_dataset |
|
|
| return dataset_dict |
|
|
| def _build_sft_datasets(self) -> DatasetDict: |
| return self._build_datasets() |
|
|
|
|
| def _build_rl_datasets(self) -> DatasetDict: |
| return self._build_datasets() |
|
|
| @classmethod |
| def _sft_preprocess(cls, example: Dict): |
| |
| format_template = r"Write an efficient and correct Python function to solve the following problem." |
| prompt_template = "Question: {prompt}\n" |
|
|
| question = example["question"].strip() |
| solution = example["solution"].strip() |
|
|
| processed_prompt = format_template + prompt_template.format(prompt=question) |
| processed_label = solution |
|
|
| text_output = { |
| "prompt": [{"role": "user", "content": processed_prompt}], |
| "completion": [{"role": "assistant", "content": processed_label}], |
| "solution": processed_label, |
| "test": example["test"].strip(), |
| "test_info": example["test_info"] |
| } |
|
|
| return text_output |
| |
| @classmethod |
| def _sft_keep_keys(cls): |
| return ["prompt", "completion", "solution", "test", "test_info"] |
|
|
|
|
|
|
