[LoRA] Cleanup LoRA unused code (#29611)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>

vllm/model_executor/models/nemotron_h.py

@@ -713,7 +713,6 @@ class NemotronHForCausalLM(
         "embed_tokens": "input_embeddings",
         "lm_head": "output_embeddings",
     }
-    embedding_padding_modules = ["lm_head"]
 
     @classmethod
     def get_mamba_state_dtype_from_config(

vllm/model_executor/models/nemotron_nas.py

@@ -387,7 +387,6 @@ class DeciLMForCausalLM(nn.Module, SupportsLoRA, SupportsPP, HasNoOps):
         "embed_tokens": "input_embeddings",
         "lm_head": "output_embeddings",
     }
-    embedding_padding_modules = ["lm_head"]
 
     # Mistral/Llama models can also be loaded with --load-format mistral
     # from consolidated.safetensors checkpoints

vllm/model_executor/models/phimoe.py

@@ -617,7 +617,6 @@ class PhiMoEForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
         "embed_tokens": "input_embeddings",
         "lm_head": "output_embeddings",
     }
-    embedding_padding_modules = ["lm_head"]
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()

vllm/model_executor/models/solar.py

@@ -426,7 +426,6 @@ class SolarForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
         "embed_tokens": "input_embeddings",
         "lm_head": "output_embeddings",
     }
-    embedding_padding_modules = ["lm_head"]
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()

vllm/model_executor/models/transformers/base.py

@@ -93,7 +93,6 @@ ALL_ATTENTION_FUNCTIONS["vllm"] = vllm_flash_attention_forward
 
 
 class Base(nn.Module, VllmModel, SupportsQuant, SupportsLoRA, SupportsPP):
-    embedding_padding_modules = ["lm_head"]
     embedding_modules = ["embed_tokens"]  # TODO transformers will have a util to get it
     hf_to_vllm_mapper = WeightsMapper(
         orig_to_new_prefix={

vllm/v1/worker/lora_model_runner_mixin.py

@@ -43,7 +43,6 @@ class LoRAModelRunnerMixin:
             vllm_config,
             device,
             model.embedding_modules,
-            model.embedding_padding_modules,
         )
         return self.lora_manager.create_lora_manager(model)