[Fix][Spec Decode] Fix llama4 draft loading with different quantization (#27136)

Signed-off-by: linzebing <[email protected]>

Author: linzebing

vllm/model_executor/models/llama4_eagle.py

@@ -60,16 +60,23 @@ def __init__(
             prefix=maybe_prefix(prefix, "embed_tokens"),
         )
 
-        self.layers = nn.ModuleList(
-            [
-                Llama4DecoderLayer(
-                    vllm_config=vllm_config,
-                    prefix=maybe_prefix(prefix, f"layers.{i + start_layer_id}"),
-                    config=self.config,
-                )
-                for i in range(self.config.num_hidden_layers)
-            ]
-        )
+        # Temporarily modify vllm_config.quant_config for draft model layers
+        original_quant_config = vllm_config.quant_config
+        vllm_config.quant_config = quant_config
+        try:
+            self.layers = nn.ModuleList(
+                [
+                    Llama4DecoderLayer(
+                        vllm_config=vllm_config,
+                        prefix=maybe_prefix(prefix, f"layers.{i + start_layer_id}"),
+                        config=self.config,
+                    )
+                    for i in range(self.config.num_hidden_layers)
+                ]
+            )
+        finally:
+            # Restore original quant_config
+            vllm_config.quant_config = original_quant_config
         self.fc = torch.nn.Linear(
             self.config.hidden_size * 2, self.config.hidden_size, bias=False
         )