Revert "[MUSA][9/N] Add FA3 attention backend support through MATE (MUSA AI Tensor Engine)" (#22002)

python/sglang/srt/server_args.py

@@ -52,7 +52,6 @@ from sglang.srt.utils.common import (
     is_hip,
     is_hopper_with_cuda_12_3,
     is_mps,
-    is_musa,
     is_no_spec_infer_or_topk_one,
     is_npu,
     is_remote_url,
@@ -2413,13 +2412,6 @@ class ServerArgs:
             if model_config.context_len > 8192:
                 self.mem_fraction_static *= 0.85
 
-        # MUSA platforms compatible backends
-        if is_musa() and self.attention_backend == "fa3":
-            logger.warning(
-                "FA3 attention backend on MUSA ignores any user-provided page_size and enforces a fixed value of 64."
-            )
-            self.page_size = 64
-
         # Other platforms backends
         if (
             self.attention_backend == "intel_amx"