| 303 | return self.tokenizer.get_lora_tokenizer(sequence.lora_request) |
| 304 | |
| 305 | def _init_tokenizer(self, **tokenizer_init_kwargs): |
| 306 | init_kwargs = dict( |
| 307 | tokenizer_id=self.model_config.tokenizer, |
| 308 | enable_lora=bool(self.lora_config), |
| 309 | max_num_seqs=self.scheduler_config.max_num_seqs, |
| 310 | max_input_length=None, |
| 311 | tokenizer_mode=self.model_config.tokenizer_mode, |
| 312 | trust_remote_code=self.model_config.trust_remote_code, |
| 313 | revision=self.model_config.tokenizer_revision) |
| 314 | init_kwargs.update(tokenizer_init_kwargs) |
| 315 | self.tokenizer = get_tokenizer_group( |
| 316 | self.parallel_config.tokenizer_pool_config, **init_kwargs) |
| 317 | |
| 318 | def _verify_args(self) -> None: |
| 319 | self.model_config.verify_with_parallel_config(self.parallel_config) |