[LoRA][1/N]Remove LoRA extra vocab (#28382)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>

vllm/model_executor/models/mamba.py

@@ -21,7 +21,6 @@ from vllm.model_executor.layers.mamba.mamba_utils import (
 )
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.vocab_parallel_embedding import (
-    DEFAULT_VOCAB_PADDING_SIZE,
     ParallelLMHead,
     VocabParallelEmbedding,
 )
@@ -110,18 +109,12 @@ class MambaModel(nn.Module):
         is_lora_enabled = bool(lora_config)
 
         self.config = config
-        lora_vocab = (
-            (lora_config.lora_extra_vocab_size * (lora_config.max_loras or 1))
-            if lora_config
-            else 0
-        )
-        self.vocab_size = config.vocab_size + lora_vocab
-        self.org_vocab_size = config.vocab_size
+
+        self.vocab_size = config.vocab_size
 
         self.embeddings = VocabParallelEmbedding(
             self.vocab_size,
             config.hidden_size,
-            org_num_embeddings=config.vocab_size,
         )
 
         self.start_layer, self.end_layer, self.layers = make_layers(
@@ -199,7 +192,7 @@ class MambaForCausalLM(
 ):
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         config = vllm_config.model_config.hf_config
-        lora_config = vllm_config.lora_config
+
         self.scheduler_config = vllm_config.scheduler_config
 
         super().__init__()
@@ -209,27 +202,17 @@ class MambaForCausalLM(
         self.backbone = MambaModel(
             vllm_config=vllm_config, prefix=maybe_prefix(prefix, "backbone")
         )
-        self.unpadded_vocab_size = config.vocab_size
-        if lora_config:
-            self.unpadded_vocab_size += lora_config.lora_extra_vocab_size
+
         if config.tie_word_embeddings:
             self.lm_head = self.backbone.embeddings
         else:
             self.lm_head = ParallelLMHead(
-                self.unpadded_vocab_size,
+                config.vocab_size,
                 config.hidden_size,
-                org_num_embeddings=config.vocab_size,
-                padding_size=DEFAULT_VOCAB_PADDING_SIZE
-                # We need bigger padding if using lora for kernel
-                # compatibility
-                if not lora_config
-                else lora_config.lora_vocab_padding_size,
                 prefix=maybe_prefix(prefix, "lm_head"),
             )
 
-        self.logits_processor = LogitsProcessor(
-            self.unpadded_vocab_size, config.vocab_size
-        )
+        self.logits_processor = LogitsProcessor(config.vocab_size)
 
         self.make_empty_intermediate_tensors = (
             self.backbone.make_empty_intermediate_tensors