Method load_weights

tensorrt_llm/_torch/modules/linear.py:2341–2353

(self,
                     weights: List[Dict],
                     allow_partial_loading: bool = False)

        return output

    def load_weights(self,
                     weights: List[Dict],
                     allow_partial_loading: bool = False):
        assert self._weights_created

        weight_mode = self.weights_loading_config.weight_mode
        if not isinstance(self.quant_method, UnquantizedLinearMethod):
            assert allow_partial_loading is False, "allow_partial_loading is only supported for non-unquantized linear methods now"
        self.quant_method.load_weights(
            self,
            weights,
            weight_mode,
            allow_partial_loading=allow_partial_loading)

    def post_load_weights(self):
        self.quant_method.post_load_weights(self)