From 53a587a18c1061f07b4dd9c1a982771802e63a72 Mon Sep 17 00:00:00 2001
From: pichangping <1337510399@qq.com>
Date: Mon, 23 Mar 2026 21:47:05 +0800
Subject: [PATCH 1/3] bugfix: Fixed the error issue when overlaying MTP and
 full decode on DSV3.1 C8.

Signed-off-by: pichangping <1337510399@qq.com>
---
 vllm_ascend/attention/mla_v1.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/vllm_ascend/attention/mla_v1.py b/vllm_ascend/attention/mla_v1.py
index aa93d460572f..b52462f580d1 100644
--- a/vllm_ascend/attention/mla_v1.py
+++ b/vllm_ascend/attention/mla_v1.py
@@ -1317,6 +1317,8 @@ def _forward_decode(
             sparse_mode = 3
             attn_mask = attn_metadata.decode.attn_mask  # type:ignore
             actual_seq_lengths = decode_meta.actual_seq_lengths_q
+            if self.fa_quant_layer:
+                dequant_scale_q_nope = dequant_scale_q_nope.view(num_tokens, self.num_heads)
         elif self.fa_quant_layer:
             attn_mask = None
             input_layout = "BSND_NBSD"
@@ -1402,7 +1404,8 @@ def _forward_decode(
                 weak_ref_tensors(softmax_lse),
             )
             if self.fa_quant_layer:
-                attn_params = attn_params + (dequant_scale_q_nope, self.fak_descale_float)  # type: ignore
+                attn_params = attn_params + (
+                    weak_ref_tensors(dequant_scale_q_nope), weak_ref_tensors(self.fak_descale_float))  # type: ignore
             else:
                 attn_params = attn_params + (None, None)  # type: ignore
 

From d5a1ac2cebadbfa732267b86e75f15c494242924 Mon Sep 17 00:00:00 2001
From: pichangping <1337510399@qq.com>
Date: Mon, 23 Mar 2026 21:58:03 +0800
Subject: [PATCH 2/3] fix CI

Signed-off-by: pichangping <1337510399@qq.com>
---
 vllm_ascend/attention/mla_v1.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/vllm_ascend/attention/mla_v1.py b/vllm_ascend/attention/mla_v1.py
index b52462f580d1..3a1790e8e838 100644
--- a/vllm_ascend/attention/mla_v1.py
+++ b/vllm_ascend/attention/mla_v1.py
@@ -1405,7 +1405,9 @@ def _forward_decode(
             )
             if self.fa_quant_layer:
                 attn_params = attn_params + (
-                    weak_ref_tensors(dequant_scale_q_nope), weak_ref_tensors(self.fak_descale_float))  # type: ignore
+                    weak_ref_tensors(dequant_scale_q_nope),
+                    weak_ref_tensors(self.fak_descale_float)
+                )  # type: ignore
             else:
                 attn_params = attn_params + (None, None)  # type: ignore
 

From fa9d361cdc8fe78ead1f1620d25dc46246f7116e Mon Sep 17 00:00:00 2001
From: pichangping <1337510399@qq.com>
Date: Mon, 23 Mar 2026 22:01:05 +0800
Subject: [PATCH 3/3] fix CI

Signed-off-by: pichangping <1337510399@qq.com>
---
 vllm_ascend/attention/mla_v1.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/vllm_ascend/attention/mla_v1.py b/vllm_ascend/attention/mla_v1.py
index 3a1790e8e838..626780cdc186 100644
--- a/vllm_ascend/attention/mla_v1.py
+++ b/vllm_ascend/attention/mla_v1.py
@@ -1406,7 +1406,7 @@ def _forward_decode(
             if self.fa_quant_layer:
                 attn_params = attn_params + (
                     weak_ref_tensors(dequant_scale_q_nope),
-                    weak_ref_tensors(self.fak_descale_float)
+                    weak_ref_tensors(self.fak_descale_float),
                 )  # type: ignore
             else:
                 attn_params = attn_params + (None, None)  # type: ignore