diff --git a/vllm/model_executor/layers/quantization/kv_cache.py b/vllm/model_executor/layers/quantization/kv_cache.py
index 4c6fcda893a0..275a1c43fdd2 100644
--- a/vllm/model_executor/layers/quantization/kv_cache.py
+++ b/vllm/model_executor/layers/quantization/kv_cache.py
@@ -125,7 +125,9 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
 
         # These are used in the final Attention.forward()
         layer._q_scale.copy_(q_scale)
-        layer._q_scale_float = q_scale
+        layer._q_scale_float = q_scale.item() if isinstance(
+            q_scale, torch.Tensor) else q_scale
+
         layer._prob_scale.copy_(prob_scale)
         if layer.kv_cache_dtype == "fp8" and (q_scale == 1.0
                                               or prob_scale == 1.0):
diff --git a/vllm/v1/attention/backends/triton_attn.py b/vllm/v1/attention/backends/triton_attn.py
index c294a5a73cbd..784912a122f6 100644
--- a/vllm/v1/attention/backends/triton_attn.py
+++ b/vllm/v1/attention/backends/triton_attn.py
@@ -361,7 +361,7 @@ def forward(
             key_cache = key_cache.view(self.fp8_dtype)
             value_cache = value_cache.view(self.fp8_dtype)
             num_tokens, num_heads, head_size = query.shape
-            assert layer._q_scale == 1.0, \
+            assert layer._q_scale_float == 1.0, \
                 "A non 1.0 q_scale is not currently supported."
             if current_platform.is_cuda():
                 # Skip Q quantization on ROCm and XPU, enable this on cuda