From 305820f1a982ed9597932778891b5da64ecccae9 Mon Sep 17 00:00:00 2001
From: jiangmengyu18 <56633611+jiangmengyu18@users.noreply.github.com>
Date: Wed, 18 Mar 2026 20:30:03 +0800
Subject: [PATCH] [Bugfix] fix bug about model type of
 qwen3_vl_8b_instruct_w8a8 (#7383)

### What this PR does / why we need it?
Adapt to the model type of Qwen3-VL-8B-Instruct-W8A8

- vLLM version: v0.17.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/4034c3d32e30d01639459edd3ab486f56993876d
---------
Signed-off-by: betta18 <jiangmengyu1@huawei.com>
Co-authored-by: betta18 <jiangmengyu1@huawei.com>
---
 .github/workflows/misc/model_list.json               |  1 +
 .github/workflows/schedule_nightly_test_a2.yaml      |  1 +
 .../models/configs/Qwen3-VL-8B-Instruct-W8A8.yaml    | 12 ++++++++++++
 vllm_ascend/quantization/modelslim_config.py         |  2 +-
 4 files changed, 15 insertions(+), 1 deletion(-)
 create mode 100644 tests/e2e/models/configs/Qwen3-VL-8B-Instruct-W8A8.yaml

diff --git a/.github/workflows/misc/model_list.json b/.github/workflows/misc/model_list.json
index 1bfd8fce..562c8601 100644
--- a/.github/workflows/misc/model_list.json
+++ b/.github/workflows/misc/model_list.json
@@ -222,6 +222,7 @@
       "vllm-ascend/Qwen3-Next-80B-A3B-Instruct-W8A8-Pruning",
       "vllm-ascend/Qwen3-Omni-30B-A3B-Thinking",
       "vllm-ascend/Qwen3-VL-8B-Instruct",
+      "vllm-ascend/Qwen3-VL-8B-Instruct-W8A8",
       "vllm-ascend/TinyLlama-1.1B-Chat-v0.3",
       "vllm-ascend/benchmark",
       "vllm-ascend/ilama-3.2-1B",
diff --git a/.github/workflows/schedule_nightly_test_a2.yaml b/.github/workflows/schedule_nightly_test_a2.yaml
index d15b3f23..f79d9496 100644
--- a/.github/workflows/schedule_nightly_test_a2.yaml
+++ b/.github/workflows/schedule_nightly_test_a2.yaml
@@ -250,6 +250,7 @@ jobs:
           - name: accuracy-group-1
             os: linux-aarch64-a2b3-1
             model_list:
+              - Qwen3-VL-8B-Instruct-W8A8
               - Qwen3-8B
               - Qwen2-Audio-7B-Instruct
               - Qwen3-8B-W8A8
diff --git a/tests/e2e/models/configs/Qwen3-VL-8B-Instruct-W8A8.yaml b/tests/e2e/models/configs/Qwen3-VL-8B-Instruct-W8A8.yaml
new file mode 100644
index 00000000..9535bd33
--- /dev/null
+++ b/tests/e2e/models/configs/Qwen3-VL-8B-Instruct-W8A8.yaml
@@ -0,0 +1,12 @@
+model_name: "vllm-ascend/Qwen3-VL-8B-Instruct-W8A8"
+hardware: "Atlas A2 Series"
+model: "vllm-vlm"
+tasks:
+- name: "mmmu_val"
+  metrics:
+  - name: "acc,none"
+    value: 0.52
+max_model_len: 8192
+batch_size: 32
+gpu_memory_utilization: 0.8
+quantization: ascend
\ No newline at end of file
diff --git a/vllm_ascend/quantization/modelslim_config.py b/vllm_ascend/quantization/modelslim_config.py
index 82b3d279..e5361dd1 100644
--- a/vllm_ascend/quantization/modelslim_config.py
+++ b/vllm_ascend/quantization/modelslim_config.py
@@ -57,7 +57,7 @@ QUANT_MODEL_PREFIX_MAPPINGS: dict[str, dict[str, str]] = {
         "language_model.lm_head.": "lm_head.",
         "language_model.model.": "model.language_model.",
     },
-    "qwen3_vl_text": {
+    "qwen3_vl": {
         "visual.": "model.visual.",
         "language_model.lm_head.": "lm_head.",
         "language_model.model.": "model.language_model.",