From d841c44e55c834650ea5dd1fc70858d90eae5f6a Mon Sep 17 00:00:00 2001
From: EX Engine <dev@enginex.ai>
Date: Mon, 10 Aug 2026 04:45:24 +0000
Subject: [PATCH] =?UTF-8?q?fix(GDN):=20dtype=20guard=20on=20=5Fix=5Fmatmul?=
 =?UTF-8?q?/=5Fix=5Fbmm=20=E2=80=94=20ixformer.matmul=20requires=20kHalf?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Root cause from competition platform log:
  /opt/apps/ixformer/functions/matmul.cu:149 'Expected input.dtype() == kHalf'
  Repeats ~80 times — every GDN layer token pass calls _ix_matmul with float32

GDN chunked delta rule uses float32 accumulation (correct for precision).
_ix_matmul was calling ixformer.matmul on float32 tensors → stderr spam.
The try/except caught it and fell back to torch.matmul, but the stderr
output floods the log and may slow down inference.

Fix: check a.dtype == torch.float16 before calling ixformer.matmul.
     Non-half tensors go directly to torch.matmul — zero stderr noise.
---
 qwen3_6_scripts/qwen3_5.py | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/qwen3_6_scripts/qwen3_5.py b/qwen3_6_scripts/qwen3_5.py
index 213b6678..fa07daea 100644
--- a/qwen3_6_scripts/qwen3_5.py
+++ b/qwen3_6_scripts/qwen3_5.py
@@ -163,8 +163,8 @@ except ImportError:
 # ---------------------------------------------------------------------------
 
 def _ix_matmul(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
-    """BI-V100 accelerated matmul via ixformer, fallback to torch."""
-    if _ix_available:
+    """BI-V100 accelerated matmul via ixformer. Only for half — ixformer rejects float32."""
+    if _ix_available and a.dtype == torch.float16:
         try:
             return _ix.matmul(a, b)
         except Exception:
@@ -172,8 +172,8 @@ def _ix_matmul(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
     return torch.matmul(a, b)
 
 def _ix_bmm(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
-    """Batched matmul — ixformer.matmul handles batched inputs."""
-    if _ix_available:
+    """Batched matmul — ixformer.matmul handles batched half inputs."""
+    if _ix_available and a.dtype == torch.float16:
         try:
             return _ix.matmul(a, b)
         except Exception:
@@ -181,8 +181,8 @@ def _ix_bmm(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
     return torch.matmul(a, b)
 
 def _ix_softmax(x: torch.Tensor, dim: int = -1) -> torch.Tensor:
-    """BI-V100 accelerated softmax via ixformer."""
-    if _ix_available:
+    """BI-V100 accelerated softmax via ixformer. Only for half."""
+    if _ix_available and x.dtype == torch.float16:
         try:
             return _ix.softmax(x, dim=dim)
         except Exception: