[Feature] Add SM103 (Blackwell Ultra) Support to vLLM (#30484)

Signed-off-by: LopezCastroRoberto <robertol.c510@gmail.com>
Signed-off-by: Roberto L. Castro <38211239+LopezCastroRoberto@users.noreply.github.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>

vllm/v1/attention/backends/mla/flashmla_sparse.py

@@ -420,7 +420,7 @@ class FlashMLASparseMetadataBuilder(AttentionMetadataBuilder[FlashMLASparseMetad
         max_num_sm_parts = int(
             max((sm_count // 2) / h_k // (cdiv(h_q // h_k, 2 * 64) * s_q), 1)
         )
-        if current_platform.is_device_capability(100):
+        if current_platform.is_device_capability_family(100):
             max_num_sm_parts *= 2
         self.tile_scheduler_metadata_buffer = torch.empty(
             # TileSchedulerMetaDataSize = 8
@@ -719,7 +719,7 @@ class FlashMLASparseImpl(MLACommonBaseImpl[FlashMLASparseMetadata]):
         self.softmax_scale = scale
         assert indexer is not None
         self.topk_indices_buffer = indexer.topk_indices_buffer
-        self.padding = 128 if current_platform.is_device_capability(100) else 64
+        self.padding = 128 if current_platform.is_device_capability_family(100) else 64
 
         if kv_cache_dtype == "fp8_ds_mla":
             # Reserve workspace during initialization