Align LoRA code between Mistral and Mixtral (fixes #2875) (#2880)

* Fix AttributeError: MixtralModel object has no attribute org_vocab_size.

* Make LoRA logic for Mistral and Mixtral the same

---------

Co-authored-by: Pernekhan Utemuratov <[email protected]>

Author: pcmoritz

vllm/model_executor/models/mixtral.py

@@ -285,15 +285,19 @@ def __init__(
         self,
         config: MixtralConfig,
         linear_method: Optional[LinearMethodBase] = None,
+        lora_config: Optional[LoRAConfig] = None,
     ) -> None:
         super().__init__()
         self.padding_idx = config.pad_token_id
-        self.vocab_size = config.vocab_size
+        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.embed_tokens = VocabParallelEmbedding(
-            config.vocab_size,
+            self.vocab_size,
             config.hidden_size,
-            org_num_embeddings=self.org_vocab_size,
+            org_num_embeddings=config.vocab_size,
         )
         self.layers = nn.ModuleList([
             MixtralDecoderLayer(config, linear_method=linear_method)
@@ -350,7 +354,9 @@ def __init__(
         super().__init__()
         self.config = config
         self.linear_method = linear_method
-        self.model = MixtralModel(config, linear_method)
+        self.model = MixtralModel(config,
+                                  linear_method,
+                                  lora_config=lora_config)
         self.unpadded_vocab_size = config.vocab_size
         if lora_config:
             self.unpadded_vocab_size += lora_config.lora_extra_vocab_size