[6/N] pass whole config to inner model (#10205)

Signed-off-by: youkaichao <youkaichao@gmail.com>

vllm/model_executor/models/mpt.py

@@ -26,7 +26,8 @@ from vllm.transformers_utils.configs.mpt import MPTConfig
 
 from .interfaces import SupportsPP
 from .utils import (is_pp_missing_parameter,
-                    make_empty_intermediate_tensors_factory, make_layers)
+                    make_empty_intermediate_tensors_factory, make_layers,
+                    maybe_prefix)
 
 
 def _get_alibi_slopes(
@@ -207,14 +208,13 @@ class MPTBlock(nn.Module):
 @support_torch_compile
 class MPTModel(nn.Module):
 
-    def __init__(
-        self,
-        config: MPTConfig,
-        cache_config: Optional[CacheConfig] = None,
-        quant_config: Optional[QuantizationConfig] = None,
-        prefix: str = "",
-    ):
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
+
+        config = vllm_config.model_config.hf_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+
         assert config.embedding_fraction == 1.0
         assert config.norm_type == "low_precision_layernorm"
 
@@ -267,20 +267,16 @@ class MPTModel(nn.Module):
 
 class MPTForCausalLM(nn.Module, SupportsPP):
 
-    def __init__(
-        self,
-        vllm_config: VllmConfig,
-        prefix: str = "",
-    ) -> None:
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
         config = vllm_config.model_config.hf_config
-        cache_config = vllm_config.cache_config
         quant_config = vllm_config.quant_config
         self.config = config
         assert config.tie_word_embeddings
         self.quant_config = quant_config
 
-        self.transformer = MPTModel(config, cache_config, quant_config)
+        self.transformer = MPTModel(vllm_config=vllm_config,
+                                    prefix=maybe_prefix(prefix, "transformer"))
         self.lm_head = self.transformer.wte
         self.logits_processor = LogitsProcessor(config.vocab_size)
         self.sampler = get_sampler()