[Feature] Move vllm/lora from SEPARATE_GROUPS to FILES in mypy config and fix type errors

tools/pre_commit/mypy.py

@@ -30,6 +30,7 @@
     "vllm/entrypoints",
     "vllm/inputs",
     "vllm/logging_utils",
+    "vllm/lora",
     "vllm/multimodal",
     "vllm/platforms",
     "vllm/transformers_utils",
@@ -47,7 +48,6 @@
     "vllm/engine",
     "vllm/executor",
     "vllm/inputs",
-    "vllm/lora",
     "vllm/model_executor",
     "vllm/plugins",
     "vllm/worker",

vllm/lora/layers/base_linear.py

@@ -152,6 +152,7 @@ def set_lora(
     def apply(
         self, x: torch.Tensor, bias: Optional[torch.Tensor] = None
     ) -> torch.Tensor:
+        assert self.base_layer.quant_method is not None
         output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
 
         # In transformers backend, x and output have extra batch dimension like

vllm/lora/layers/column_parallel_linear.py

@@ -35,6 +35,7 @@ def _mcp_apply(x, bias, layer: "ColumnParallelLinearWithLoRA"):
     if layer.lora_bias_stacked is not None:
         assert layer.n_slices == len(layer.lora_bias_stacked)
 
+    assert layer.base_layer.quant_method is not None
     output = layer.base_layer.quant_method.apply(layer.base_layer, x, bias)
 
     x = x.view(-1, x.shape[-1])

vllm/lora/layers/row_parallel_linear.py

@@ -136,6 +136,7 @@ def slice_bias(self, bias: torch.Tensor) -> torch.Tensor:
     def apply(
         self, x: torch.Tensor, bias: Optional[torch.Tensor] = None
     ) -> torch.Tensor:
+        assert self.base_layer.quant_method is not None
         output = self.base_layer.quant_method.apply(self.base_layer, x)
 
         x = x.view(-1, x.shape[-1])

vllm/lora/models.py

@@ -252,6 +252,7 @@ def check_unexpected_modules(modules: dict):
             from tensorizer import TensorDeserializer
 
             tensorizer_config = TensorizerConfig(**tensorizer_config_dict)
+            assert tensorizer_config.tensorizer_dir is not None
             lora_tensor_path = os.path.join(
                 tensorizer_config.tensorizer_dir, "adapter_model.tensors"
             )
@@ -356,6 +357,7 @@ def __init__(
             vocab_size: the vocab size of the model.
             lora_config: the LoRA configuration.
         """
+        assert isinstance(model, nn.Module)
         self.model: SupportsLoRA = model
         self._registered_adapters: dict[int, LoRAModel] = {}
         # Dict instead of a set for compatibility with LRUCache.
@@ -394,13 +396,14 @@ def __init__(
         # Dict instead of a set for compatibility with LRUCache.
         self._last_mapping: Optional[LoRAMapping] = None
         self._create_lora_modules()
-        self.model.lora_manager = self
+        setattr(self.model, "lora_manager", self)
 
     def __len__(self) -> int:
         return len(self._registered_adapters)
 
     @property
     def capacity(self) -> int:
+        assert self.lora_config.max_cpu_loras is not None
         return self.lora_config.max_cpu_loras
 
     @property
@@ -503,7 +506,7 @@ def _parent_module(module_name: str) -> str:
             #  - given an input 'x' return ''
             return module_name.rpartition(".")[0]
 
-        for module_name, module in self.model.named_modules(remove_duplicate=False):
+        for module_name, module in self.model.named_modules(remove_duplicate=False):  # type: ignore[attr-defined]
             if isinstance(module, PPMissingLayer):
                 continue
             if not self._match_target_modules(module_name):
@@ -527,7 +530,7 @@ def _parent_module(module_name: str) -> str:
                     self.lora_slots,
                     self.lora_config,
                     packed_moduled_lst,
-                    self.model.config,
+                    self.model.config,  # type: ignore[attr-defined]
                 ),
             )
 
@@ -540,7 +543,7 @@ def _parent_module(module_name: str) -> str:
                         f"{parent_module}.{logits_processor_module_name}"
                     )
 
-                logits_processor_module = self.model.get_submodule(
+                logits_processor_module = self.model.get_submodule(  # type: ignore[attr-defined]
                     logits_processor_module_name
                 )
 
@@ -552,7 +555,7 @@ def _parent_module(module_name: str) -> str:
                         module,
                         self.lora_slots,
                         self.lora_config,
-                        self.model.config,
+                        self.model.config,  # type: ignore[attr-defined]
                     ),
                 )
 
@@ -580,7 +583,7 @@ def create_dummy_lora(
     ) -> LoRAModel:
         """Create zero-initialized LoRAModel for warmup."""
         model = LoRAModel(lora_id, rank, {})
-        for module_name, module in self.model.named_modules():
+        for module_name, module in self.model.named_modules():  # type: ignore[attr-defined]
             bias_enabled = self.lora_config.bias_enabled
             if (
                 not self._match_target_modules(module_name)
@@ -663,7 +666,7 @@ def _filter_unsupported_mm_module(self, module_name: str) -> bool:
         be filtered out.
         """
         if self.supports_mm:
-            module_mapping: MultiModelKeys = self.model.get_mm_mapping()
+            module_mapping: MultiModelKeys = self.model.get_mm_mapping()  # type: ignore[attr-defined]
             prefix_lst = module_mapping.connector + module_mapping.tower_model
             return any([module_name.startswith(prefix) for prefix in prefix_lst])
         return False

vllm/lora/peft_helper.py

@@ -91,6 +91,7 @@ def from_local_dir(
             tensorizer_args = tensorizer_config._construct_tensorizer_args()
             from tensorizer.stream_io import open_stream
 
+            assert tensorizer_config.tensorizer_dir is not None
             lora_config_path = os.path.join(
                 tensorizer_config.tensorizer_dir, "adapter_config.json"
             )

vllm/lora/utils.py

@@ -132,11 +132,13 @@ def parse_fine_tuned_lora_name(
     # mapping correctly.
     if name.startswith("base_model.model."):
         name = name.replace("base_model.model.", "")
-        name = weights_mapper._map_name(name) if weights_mapper else name
+        mapped_name = weights_mapper._map_name(name) if weights_mapper else name
+        name = mapped_name if mapped_name is not None else name
         # recover the prefix `base_model.model.`
         name = "base_model.model." + name
     else:
-        name = weights_mapper._map_name(name) if weights_mapper else name
+        mapped_name = weights_mapper._map_name(name) if weights_mapper else name
+        name = mapped_name if mapped_name is not None else name
 
     # In some situations, we may not start with `base_model.model.`.
     # If we don't (e.g., ibm-granite/granite-speech-3.3-8b),

vllm/lora/worker_manager.py

@@ -7,6 +7,7 @@
 import torch
 
 from vllm.config import VllmConfig
+from vllm.config.lora import LoRAConfig
 from vllm.logger import init_logger
 from vllm.lora.models import (
     LoRAModel,
@@ -46,7 +47,9 @@ def __init__(
             vllm_config.scheduler_config.max_num_batched_tokens
         )
         self.vocab_size = vllm_config.model_config.get_vocab_size()
-        self.lora_config = vllm_config.lora_config
+        lora_config = vllm_config.lora_config
+        assert lora_config is not None
+        self.lora_config: LoRAConfig = lora_config
 
         # Use get_text_config() in case of multimodal models
         text_config = vllm_config.model_config.hf_config.get_text_config()