[Bugfix] Voxtral on Blackwell GPUs (RTX 50 series) (#21077)

Signed-off-by: hax0r31337 <liulihaocaiqwq@gmail.com>

vllm/attention/layer.py

@@ -16,6 +16,7 @@ from vllm.distributed.kv_transfer import (get_kv_transfer_group,
                                           has_kv_transfer_group,
                                           is_v1_kv_transfer_group)
 from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.logger import init_logger
 from vllm.model_executor.layers.linear import UnquantizedLinearMethod
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig)
@@ -23,6 +24,34 @@ from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
 from vllm.platforms import _Backend, current_platform
 from vllm.utils import direct_register_custom_op
 
+logger = init_logger(__name__)
+USE_XFORMERS_OPS = None
+
+
+def check_xformers_availability():
+    global USE_XFORMERS_OPS
+    if USE_XFORMERS_OPS is not None:
+        return USE_XFORMERS_OPS
+
+    if current_platform.is_cuda() and current_platform.has_device_capability(
+            100):
+        # Xformers FA is not compatible with B200
+        USE_XFORMERS_OPS = False
+    else:
+        try:
+            from importlib.util import find_spec
+
+            find_spec("xformers.ops")
+            USE_XFORMERS_OPS = True
+        except ImportError:
+            USE_XFORMERS_OPS = False
+
+    # the warning only needs to be shown once
+    if not USE_XFORMERS_OPS:
+        logger.warning("Xformers is not available, falling back.")
+
+    return USE_XFORMERS_OPS
+
 
 class Attention(nn.Module):
     """Attention layer.
@@ -314,6 +343,10 @@ class MultiHeadAttention(nn.Module):
                 _Backend.TORCH_SDPA, _Backend.XFORMERS, _Backend.PALLAS_VLLM_V1
             } else _Backend.TORCH_SDPA
 
+        if (self.attn_backend == _Backend.XFORMERS
+                and not check_xformers_availability()):
+            self.attn_backend = _Backend.TORCH_SDPA
+
     def forward(
         self,
         query: torch.Tensor,