[Bugfix] Revert "Zero-init MLA attention output buffers to prevent NaN from CUDA graph padding (#37442)"

This reverts commit ef2c4f778d.

The zero-init workaround is unnecessary — the NaN was caused by a
different issue (int64 expert IDs in the routing simulator). Reverting
to restore the original torch.empty allocation which avoids the
overhead of pre-allocated zero-init buffers.

Signed-off-by: Elvir Crncevic <elvircrn@gmail.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

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

@@ -162,11 +162,6 @@ class CutlassMLAImpl(MLACommonImpl[MLACommonMetadata]):
         # Share workspace buffer across all executions
         self._workspace = g_sm100_workspace
 
-        # Pre-allocated output buffer, lazily sized on first call.
-        # Zero-init once to prevent NaN in padding slots (seq_lens=0)
-        # from contaminating downstream per-tensor reductions.
-        self._decode_out: torch.Tensor | None = None
-
     def _sm100_cutlass_mla_decode(
         self,
         q_nope: torch.Tensor,
@@ -223,15 +218,7 @@ class CutlassMLAImpl(MLACommonImpl[MLACommonMetadata]):
             if is_quantized_kv_cache(self.kv_cache_dtype)
             else q_nope.dtype
         )
-        # Reuse pre-allocated zero-init output buffer to avoid a memset
-        # kernel on every CUDA graph replay.
-        if (
-            self._decode_out is None
-            or self._decode_out.shape[0] < B_q
-            or self._decode_out.dtype != dtype
-        ):
-            self._decode_out = q_nope.new_zeros((B_q, MAX_HEADS, D_latent), dtype=dtype)
-        out = self._decode_out[:B_q]
+        out = q_nope.new_empty((B_q, MAX_HEADS, D_latent), dtype=dtype)
         lse = (
             torch.empty((B_q, MAX_HEADS), dtype=torch.float32, device=q_nope.device)
             if self.need_to_return_lse_for_decode

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

@@ -21,7 +21,6 @@ from vllm.v1.attention.backend import (
     AttentionLayer,
     AttentionType,
     MultipleOf,
-    is_quantized_kv_cache,
 )
 from vllm.v1.attention.backends.utils import KVCacheLayoutType
 
@@ -152,11 +151,6 @@ class FlashInferMLAImpl(MLACommonImpl[MLACommonMetadata]):
         self.bmm1_scale: float | None = None
         self.bmm2_scale: float | None = None
 
-        # Pre-allocated output buffer, lazily sized on first call.
-        # Zero-init once to prevent NaN in padding slots (seq_lens=0)
-        # from contaminating downstream per-tensor reductions.
-        self._decode_out: torch.Tensor | None = None
-
     def forward_mqa(
         self,
         q: torch.Tensor | tuple[torch.Tensor, torch.Tensor],
@@ -192,37 +186,6 @@ class FlashInferMLAImpl(MLACommonImpl[MLACommonMetadata]):
             if self.kv_cache_dtype.startswith("fp8"):
                 self.bmm2_scale *= layer._k_scale_float
 
-        # Reuse pre-allocated zero-init output buffer to avoid a memset
-        # kernel on every CUDA graph replay.
-        # q is 4D: (batch, q_len_per_req, num_heads, head_dim)
-        # FlashInfer has a bug where out= validation hardcodes 3D shape
-        # (batch, num_heads, kv_lora_rank), but the kernel writes 4D
-        # (batch, q_len, num_heads, kv_lora_rank) when q_len > 1.
-        # So we can only pass out= for single-token decode (q_len == 1).
-        # For q_len > 1, we zero padding slots after the kernel returns.
-        # TODO: upstream fix to FlashInfer
-        B, q_len_per_req = q.shape[0], q.shape[1]
-        out_kwargs: dict[str, torch.Tensor] = {}
-        if q_len_per_req == 1:
-            dtype = (
-                torch.bfloat16
-                if is_quantized_kv_cache(self.kv_cache_dtype)
-                else q.dtype
-            )
-            if (
-                self._decode_out is None
-                or self._decode_out.shape[0] < B
-                or self._decode_out.dtype != dtype
-            ):
-                self._decode_out = torch.zeros(
-                    B,
-                    q.shape[2],
-                    self.kv_lora_rank,
-                    dtype=dtype,
-                    device=q.device,
-                )
-            out_kwargs["out"] = self._decode_out[:B]
-
         o = trtllm_batch_decode_with_kv_cache_mla(
             query=q,
             kv_cache=kv_c_and_k_pe_cache.unsqueeze(1),
@@ -235,15 +198,8 @@ class FlashInferMLAImpl(MLACommonImpl[MLACommonMetadata]):
             max_seq_len=attn_metadata.max_seq_len,
             bmm1_scale=self.bmm1_scale,
             bmm2_scale=self.bmm2_scale,
-            **out_kwargs,
         )
 
-        # For q_len > 1, we can't pass out= so we work around by zeroing padding slots
-        if not out_kwargs:
-            num_real = attn_metadata.num_decodes
-            if num_real < o.shape[0]:
-                o[num_real:] = 0
-
         # Flatten the output for consistent shape
         o = o.view(-1, o.shape[-2], o.shape[-1])