Dataset for supervised fine-tuning.
| 167 | |
| 168 | |
| 169 | class SupervisedDataset(Dataset): |
| 170 | """Dataset for supervised fine-tuning.""" |
| 171 | |
| 172 | def __init__(self, raw_data, tokenizer: transformers.PreTrainedTokenizer, template="tool-llama"): |
| 173 | super(SupervisedDataset, self).__init__() |
| 174 | |
| 175 | rank0_print("Formatting inputs...") |
| 176 | sources = [example["conversations"] for example in raw_data] |
| 177 | self.template = template |
| 178 | data_dict = preprocess(sources, tokenizer, self.template) |
| 179 | self.input_ids = data_dict["input_ids"] |
| 180 | self.labels = data_dict["labels"] |
| 181 | self.attention_mask = data_dict["attention_mask"] |
| 182 | |
| 183 | def __len__(self): |
| 184 | return len(self.input_ids) |
| 185 | |
| 186 | def __getitem__(self, i) -> Dict[str, torch.Tensor]: |
| 187 | return dict( |
| 188 | input_ids=self.input_ids[i], |
| 189 | labels=self.labels[i], |
| 190 | attention_mask=self.attention_mask[i], |
| 191 | ) |
| 192 | |
| 193 | |
| 194 | class LazySupervisedDataset(Dataset): |
nothing calls this directly
no outgoing calls
no test coverage detected