[Bugfix] Work around FlashInfer unified transport issue on GB (#20039)

python/sglang/srt/environ.py

@@ -342,6 +342,9 @@ class Envs:
     # Default to the pick from flashinfer
     SGLANG_FLASHINFER_FP4_GEMM_BACKEND = EnvStr("")
     SGLANG_FLASHINFER_WORKSPACE_SIZE = EnvInt(384 * 1024 * 1024)
+    # TODO(mmangkad): Remove this once the FlashInfer unified allreduce-fusion
+    # transport issue on GB200/GB300 platforms is fixed and verified resolved.
+    SGLANG_FLASHINFER_FORCE_POSIX_FD_TRANSPORT = EnvBool(None)
 
     # Triton
     SGLANG_TRITON_DECODE_ATTN_STATIC_KV_SPLITS = EnvBool(False)

python/sglang/srt/layers/flashinfer_comm_fusion.py

@@ -1,4 +1,6 @@
+import contextlib
 import logging
+import platform
 from typing import Optional, Tuple
 
 import torch
@@ -11,6 +13,7 @@ from sglang.srt.distributed import (
     get_moe_tensor_parallel_rank,
     get_moe_tensor_parallel_world_size,
 )
+from sglang.srt.environ import envs
 from sglang.srt.utils import is_flashinfer_available
 from sglang.srt.utils.custom_op import register_custom_op
 
@@ -19,6 +22,73 @@ logger = logging.getLogger(__name__)
 _flashinfer_comm = None
 _workspace_manager = None
 _flashinfer_allreduce_unavailable = False
+_posix_transport_override_logged = False
+
+
+def _should_force_posix_fd_transport() -> bool:
+    force_posix_env = envs.SGLANG_FLASHINFER_FORCE_POSIX_FD_TRANSPORT.get()
+    if force_posix_env is not None:
+        return force_posix_env
+
+    machine = platform.machine().lower()
+    if machine not in ("aarch64", "arm64"):
+        return False
+
+    if not torch.cuda.is_available():
+        return False
+
+    try:
+        major, _minor = torch.cuda.get_device_capability(torch.cuda.current_device())
+    except Exception as e:
+        logger.debug("Failed to get CUDA device capability: %s", e)
+        return False
+
+    return major == 10
+
+
+@contextlib.contextmanager
+def _flashinfer_posix_fd_transport_override_if_needed():
+    # TODO(mmangkad): Remove this temporary override once the
+    # FlashInfer unified allreduce-fusion transport issue on
+    # GB200/GB300 platforms is fixed and verified resolved.
+    global _posix_transport_override_logged
+
+    if not _should_force_posix_fd_transport():
+        yield
+        return
+
+    try:
+        import flashinfer.comm.mnnvl as flashinfer_mnnvl
+    except Exception as e:
+        logger.debug(
+            "Failed to import flashinfer.comm.mnnvl for transport override: %s", e
+        )
+        yield
+        return
+
+    original_checker = getattr(flashinfer_mnnvl, "is_mnnvl_fabric_supported", None)
+    if original_checker is None:
+        yield
+        return
+
+    if not _posix_transport_override_logged:
+        logger.warning(
+            "Applying FlashInfer transport workaround: forcing PosixFD "
+            "symmetric-memory handle exchange on aarch64 + sm10x to avoid "
+            "known data corruption with Fabric handle exchange on GB systems. "
+            "Set SGLANG_FLASHINFER_FORCE_POSIX_FD_TRANSPORT=0 to disable."
+        )
+        _posix_transport_override_logged = True
+
+    def _always_disable_fabric(_device_idx: int) -> bool:
+        return False
+
+    flashinfer_mnnvl.is_mnnvl_fabric_supported = _always_disable_fabric
+    try:
+        yield
+    finally:
+        flashinfer_mnnvl.is_mnnvl_fabric_supported = original_checker
+
 
 if is_flashinfer_available():
     try:
@@ -74,15 +144,16 @@ class FlashInferWorkspaceManager:
 
         self.cleanup()
         try:
-            self.workspace = _flashinfer_comm.create_allreduce_fusion_workspace(
+            with _flashinfer_posix_fd_transport_override_if_needed():
-                backend="trtllm",
+                self.workspace = _flashinfer_comm.create_allreduce_fusion_workspace(
-                world_size=world_size,
+                    backend="trtllm",
-                rank=rank,
+                    world_size=world_size,
-                max_token_num=max_token_num,
+                    rank=rank,
-                hidden_dim=hidden_dim,
+                    max_token_num=max_token_num,
-                dtype=dtype,
+                    hidden_dim=hidden_dim,
-                force_oneshot_support=bool(use_oneshot),
+                    dtype=dtype,
-            )
+                    force_oneshot_support=bool(use_oneshot),
+                )
         except Exception as e:
             global _flashinfer_allreduce_unavailable
             _flashinfer_allreduce_unavailable = True