From 021f95d01ef6db9614c1d17a0fe6e20706ef4000 Mon Sep 17 00:00:00 2001
From: Aaron Ang <67321817+aaron-ang@users.noreply.github.com>
Date: Fri, 13 Jun 2025 16:51:57 -0700
Subject: [PATCH] Only support int8 dtype for quant operators

---
 backends/xnnpack/partition/config/xnnpack_config.py | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/backends/xnnpack/partition/config/xnnpack_config.py b/backends/xnnpack/partition/config/xnnpack_config.py
index df6067a7d68..e0221c27d44 100644
--- a/backends/xnnpack/partition/config/xnnpack_config.py
+++ b/backends/xnnpack/partition/config/xnnpack_config.py
@@ -10,12 +10,13 @@
 from typing import List, Optional
 
 import torch
+from torch.export import ExportedProgram
+from executorch.backends.xnnpack.utils.quant_utils import is_dynamic_qdq
 from executorch.exir.backend.canonical_partitioners.config_partitioner import (
     format_target_name,
     PartitionerConfig,
 )
 from executorch.exir.backend.utils import WhyNoPartition
-from torch.export import ExportedProgram
 
 logger = logging.getLogger(__name__)
 why = WhyNoPartition(logger=logger)
@@ -220,9 +221,12 @@ def _check_node_has_valid_dtype(self, node):
         valid_dtypes = {
             torch.float32,
             torch.float16,
-            torch.int8,
             torch.qint8,
         }
+        # Only allow int8 for quantization operations
+        if is_dynamic_qdq(node):
+            valid_dtypes.add(torch.int8)
+
         if (
             node.op != "placeholder"
             and node.op != "call_function"