From 3e01f3a533d3135049b3a445617e85ce05e0ab47 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=D0=90=D1=80=D1=82=D0=B5=D0=BC=20=D0=A1=D0=B0=D0=B2=D0=BA?=
 =?UTF-8?q?=D0=B8=D0=BD?= <58187114+OrangeRedeng@users.noreply.github.com>
Date: Sat, 20 Dec 2025 09:53:46 +0300
Subject: [PATCH] [NPU] [BUGFIX] [CRITICAL!] Fix NPU inference
 (torch_npu._npu_reshape_and_cache() crash) (#15484)

---
 .../hardware_backend/npu/memory_pool_npu.py   | 40 +++++++++----------
 1 file changed, 20 insertions(+), 20 deletions(-)

diff --git a/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py b/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py
index 41fb2aae2..4968c03e8 100644
--- a/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py
+++ b/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py
@@ -34,18 +34,18 @@ class NPUMHATokenToKVPool(MHATokenToKVPool):
     ):
         self.use_fia = get_bool_env_var("ASCEND_USE_FIA", "False")
         super().__init__(
-            size,
-            page_size,
-            dtype,
-            head_num,
-            head_dim,
-            layer_num,
-            device,
-            enable_memory_saver,
-            start_layer,
-            end_layer,
-            enable_alt_stream,
-            enable_kv_cache_copy,
+            size=size,
+            page_size=page_size,
+            dtype=dtype,
+            head_num=head_num,
+            head_dim=head_dim,
+            layer_num=layer_num,
+            device=device,
+            enable_memory_saver=enable_memory_saver,
+            start_layer=start_layer,
+            end_layer=end_layer,
+            enable_alt_stream=enable_alt_stream,
+            enable_kv_cache_copy=enable_kv_cache_copy,
         )
 
     def _create_buffers(self):
@@ -181,14 +181,14 @@ class NPUMLATokenToKVPool(MLATokenToKVPool):
         end_layer: Optional[int] = None,
     ):
         super(MLATokenToKVPool, self).__init__(
-            size,
-            page_size,
-            dtype,
-            layer_num,
-            device,
-            enable_memory_saver,
-            start_layer,
-            end_layer,
+            size=size,
+            page_size=page_size,
+            dtype=dtype,
+            layer_num=layer_num,
+            device=device,
+            enable_memory_saver=enable_memory_saver,
+            start_layer=start_layer,
+            end_layer=end_layer,
         )
 
         self.kv_lora_rank = kv_lora_rank