[Misc][Refactor] Generalize linear_method to be quant_method (#4373)

vllm/model_executor/models/orion.py

@@ -13,11 +13,12 @@ from transformers import PretrainedConfig
 from vllm.attention import Attention, AttentionMetadata
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import SiluAndMul
-from vllm.model_executor.layers.linear import (LinearMethodBase,
-                                               MergedColumnParallelLinear,
+from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
                                                QKVParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
+from vllm.model_executor.layers.quantization.base_config import (
+    QuantizationConfig)
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.sampler import Sampler
 from vllm.model_executor.layers.vocab_parallel_embedding import (
@@ -34,17 +35,17 @@ class OrionMLP(nn.Module):
         hidden_size: int,
         intermediate_size: int,
         hidden_act: str,
-        linear_method: Optional[LinearMethodBase] = None,
+        quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
         self.gate_up_proj = MergedColumnParallelLinear(
             hidden_size, [intermediate_size] * 2,
             bias=False,
-            linear_method=linear_method)
+            quant_config=quant_config)
         self.down_proj = RowParallelLinear(intermediate_size,
                                            hidden_size,
                                            bias=False,
-                                           linear_method=linear_method)
+                                           quant_config=quant_config)
         if hidden_act != "silu":
             raise ValueError(f"Unsupported activation: {hidden_act}. "
                              "Only silu is supported for now.")
@@ -67,7 +68,7 @@ class OrionAttention(nn.Module):
         rope_theta: float = 10000,
         rope_scaling: Optional[Dict[str, Any]] = None,
         max_position_embeddings: int = 8192,
-        linear_method: Optional[LinearMethodBase] = None,
+        quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
         self.hidden_size = hidden_size
@@ -98,13 +99,13 @@ class OrionAttention(nn.Module):
             self.total_num_heads,
             self.total_num_kv_heads,
             bias=False,
-            linear_method=linear_method,
+            quant_config=quant_config,
         )
         self.o_proj = RowParallelLinear(
             self.total_num_heads * self.head_dim,
             hidden_size,
             bias=False,
-            linear_method=linear_method,
+            quant_config=quant_config,
         )
 
         self.rotary_emb = get_rope(
@@ -139,7 +140,7 @@ class OrionDecoderLayer(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        linear_method: Optional[LinearMethodBase] = None,
+        quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
         self.hidden_size = config.hidden_size
@@ -154,13 +155,13 @@ class OrionDecoderLayer(nn.Module):
             rope_theta=rope_theta,
             rope_scaling=rope_scaling,
             max_position_embeddings=max_position_embeddings,
-            linear_method=linear_method,
+            quant_config=quant_config,
         )
         self.mlp = OrionMLP(
             hidden_size=self.hidden_size,
             intermediate_size=config.intermediate_size,
             hidden_act=config.hidden_act,
-            linear_method=linear_method,
+            quant_config=quant_config,
         )
 
         self.input_layernorm = nn.LayerNorm(config.hidden_size,
@@ -201,7 +202,7 @@ class OrionModel(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        linear_method: Optional[LinearMethodBase] = None,
+        quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
         self.config = config
@@ -212,7 +213,7 @@ class OrionModel(nn.Module):
             config.hidden_size,
         )
         self.layers = nn.ModuleList([
-            OrionDecoderLayer(config, linear_method)
+            OrionDecoderLayer(config, quant_config)
             for _ in range(config.num_hidden_layers)
         ])
         self.norm = nn.LayerNorm(config.hidden_size, eps=config.rms_norm_eps)
@@ -244,12 +245,12 @@ class OrionForCausalLM(nn.Module):
     def __init__(
         self,
         config: PretrainedConfig,
-        linear_method: Optional[LinearMethodBase] = None,
+        quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
         self.config = config
-        self.linear_method = linear_method
-        self.model = OrionModel(config, linear_method)
+        self.quant_config = quant_config
+        self.model = OrionModel(config, quant_config)
         self.lm_head = ParallelLMHead(config.vocab_size, config.hidden_size)
         self.logits_processor = LogitsProcessor(config.vocab_size)
         self.sampler = Sampler()