[Kernel][CPU] CPU MLA (#14744)

Signed-off-by: Thien Tran <gau.nernst@yahoo.com.sg>

vllm/platforms/cpu.py

@@ -37,6 +37,9 @@ class CpuPlatform(Platform):
                              use_mla: bool) -> str:
         if selected_backend and selected_backend != _Backend.TORCH_SDPA:
             logger.info("Cannot use %s backend on CPU.", selected_backend)
+        if use_mla:
+            logger.info("Using CPU MLA backend.")
+            return "vllm.attention.backends.cpu_mla.CPUMLABackend"
         logger.info("Using Torch SDPA backend.")
         return "vllm.attention.backends.torch_sdpa.TorchSDPABackend"
 
@@ -129,9 +132,6 @@ class CpuPlatform(Platform):
         # Disable torch async compiling which won't work with daemonic processes
         os.environ["TORCHINDUCTOR_COMPILE_THREADS"] = "1"
 
-        # MLA attention is not supported
-        os.environ["VLLM_MLA_DISABLE"] = "1"
-
         # Intel OpenMP setting
         ld_prealod_str = os.getenv("LD_PRELOAD", "")
         if "libiomp5.so" in ld_prealod_str: