Weight loading in vLLM

Fixed a weight loading error. It was reporting There is no module or parameter named sth at weight loading. It took me couple of days to root cause this issue. One step closer the goal.

1 Weight loading workflow

## worker/gpu_worker.py
class Worker(Workerbase):
  def load_model(self) -> None:
    ...
    self.model_runner.load_model()
## worker/gpu_model_runner.py
class GPUModelRunner()
  def load_model(self) -> None:
    ...
    model_loader = get_model_loader(self.load_config)
    self.model = model_loader.load_model(
                    vllm_config=self.vllm_config,
                    model_config=self.model_config)
    ...
## model_loader/base_loader.py
class BaseModelLoader(ABC):
  def load_model(self, vllm_config: VllmConfig,
                  model_config: ModelConfig) -> nn.Module:
      ...
      self.load_weights(model, model_config)
      process_weights_after_loading(model, model_config, target_device)
      ...
## model_loader/default_loader.py
class DefaultModelLoader(BaseModelLoader):
  def load_weights(self, model: nn.Module,
                    model_config: ModelConfig) -> None:
      # List all the weights needs to be loaded
      weights_to_load = {name for name, _ in model.named_parameters()}
      loaded_weights = model.load_weights(
          self.get_all_weights(model_config, model))
      ...
## models/model_name.py
def load_weights(self, weights: Iterable[tuple[str,
                torch.Tensor]]) -> set[str]:
    loader = AutoWeightsLoader(self)
    return loader.load_weights(weights)
## models/utils.py
class AutoWeightsLoader:
  def load_weights(
          self,
          weights: Iterable[tuple[str, torch.Tensor]],
          *,
          mapper: Optional[WeightsMapper] = None,
      ) -> set[str]:
      ...
      autoloaded_weights = set(self._load_module("", self.module, weights))
      ...
  def _load_module(
        self,
        base_prefix: str,
        module: nn.Module,
        weights: Iterable[tuple[str, torch.Tensor]],
    ) -> Iterable[str]:
      # call recursively
      ...
      msg = (f"There is no module or parameter named '{prefix}' "
             f"in {type(self.module).__name__}")
      raise ValueError(msg)

2 Skip loading options

When we overriding load_weights, we can skip some attributes.
In this examples, norm_mean and norm_std are registerd_buffer instead of modules. So skip them could fix the bug.

def def load_weights(...)
  skip_substrs = ["norm_mean", "norm_std"]
  loader = AutoWeightsLoader(self, skip_substrs=skip_substrs)

The next bug is `hf_pr