Method load_adapter

optillm/inference.py:1147–1186

Load a LoRA adapter with enhanced caching

(self, base_model: PreTrainedModel, adapter_id: str)

            return False

    def load_adapter(self, base_model: PreTrainedModel, adapter_id: str) -> PreTrainedModel:
        """Load a LoRA adapter with enhanced caching"""
        model_key = base_model.config._name_or_path
        
        def _load_adapter():
            logger.info(f"Loading LoRA adapter: {adapter_id}")
            
            if not self.validate_adapter(adapter_id):
                error_msg = f"Adapter {adapter_id} not found or is not compatible"
                logger.error(error_msg)
                raise ValueError(error_msg)
            
            try:
                adapter_name = self._get_adapter_name(adapter_id)
                
                config = PeftConfig.from_pretrained(
                    adapter_id,
                    trust_remote_code=True,
                    token=os.getenv("HF_TOKEN")
                )
                
                model = base_model
                model.add_adapter(
                    config,
                    adapter_name = adapter_name,
                )
                
                if model not in self.loaded_adapters:
                    self.loaded_adapters[model] = []
                if adapter_id not in self.loaded_adapters[model]:
                    self.loaded_adapters[model].append(adapter_id)
                
                return model
                
            except Exception as e:
                error_msg = f"Failed to load adapter {adapter_id}: {str(e)}"
                logger.error(error_msg)
                raise RuntimeError(error_msg) from e

        return self.cache_manager.get_or_load_adapter(model_key, adapter_id, _load_adapter)

    def set_active_adapter(self, model: PeftModel, adapter_id: str = None) -> bool:
        """Set a specific adapter as active with error handling"""