[Misc] Support quantization of MllamaForCausalLM (#8822)

vllm/model_executor/models/mllama.py

@@ -624,6 +624,7 @@ class MllamaTextCrossAttention(nn.Module):
         self,
         config: Optional[config_mllama.MllamaTextConfig] = None,
         layer_idx: Optional[int] = None,
+        quant_config: Optional[QuantizationConfig] = None,
     ):
         super().__init__()
         self.config = config
@@ -648,12 +649,14 @@ class MllamaTextCrossAttention(nn.Module):
             self.num_heads,
             self.num_key_value_heads,
             bias=False,
+            quant_config=quant_config,
         )
         self.o_proj = RowParallelLinear(
             self.num_heads * self.head_dim,
             self.hidden_size,
             bias=False,
             input_is_parallel=True,
+            quant_config=quant_config,
         )
         # vllm.model_executor.layers.layernorm.RMSNorm has precision issue,
         # use huggingface's instead
@@ -708,13 +711,15 @@ class MllamaCrossAttentionDecoderLayer(torch.nn.Module):
     """Cross-attention transformer block with tanh-gated attention
     and feedforward."""
 
-    def __init__(self, config: config_mllama.MllamaTextConfig, layer_idx: int) \
+    def __init__(self, config: config_mllama.MllamaTextConfig, layer_idx: int,
+                 quant_config: Optional[QuantizationConfig]) \
         -> None:
         super().__init__()
         self.layer_idx = layer_idx
         self.cross_attn = MllamaTextCrossAttention(
             config=config,
             layer_idx=layer_idx,
+            quant_config=quant_config,
         )
 
         self.input_layernorm = RMSNorm(config.hidden_size,
@@ -725,6 +730,7 @@ class MllamaCrossAttentionDecoderLayer(torch.nn.Module):
             hidden_size=config.hidden_size,
             intermediate_size=config.intermediate_size,
             hidden_act=config.hidden_act,
+            quant_config=quant_config,
         )
         self.post_attention_layernorm = RMSNorm(config.hidden_size,
                                                 eps=config.rms_norm_eps)
@@ -780,7 +786,8 @@ class MllamaTextModel(nn.Module):
         for layer_idx in range(config.num_hidden_layers):
             if layer_idx in self.cross_attention_layers:
                 layers.append(
-                    MllamaCrossAttentionDecoderLayer(config, layer_idx))
+                    MllamaCrossAttentionDecoderLayer(
+                        config, layer_idx, quant_config=quant_config))
             else:
                 # TODO: force LlamaDecoderLayer to config.attention_bias=False
                 layers.append(