(self, raw_data, tokenizer: transformers.PreTrainedTokenizer, template="tool-llama")
| 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) |
no test coverage detected