EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[TEST] Add Qwen3-32b-w8a8 acc/perf A2/A3 test (#3541)

Browse Source 
### What this PR does / why we need it?
This PR Qwen3-32b-w8a8 acc/perf 8 cases on A2 and A3, we need test them
daily.

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
by running the test


- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: jiangyunfan1 <jiangyunfan1@h-partners.com>
Signed-off-by: wangli <wangli858794774@gmail.com>
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
Signed-off-by: root <root@hostname-2pbfv.foreman.pxe>
Co-authored-by: wangli <wangli858794774@gmail.com>
Co-authored-by: Yikun Jiang <yikunkero@gmail.com>
This commit is contained in:
jiangyunfan1
2025-10-21 17:34:48 +08:00
committed by GitHub
parent ec1d2b5c04
commit 80b8df881f
6 changed files with 307 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
tools/aisbench.py
View File

@@ -101,6 +101,9 @@ class AisbenchRunner:
if self.task_type == "performance":
conf_path = os.path.join(DATASET_CONF_DIR,
f'{self.dataset_conf}.py')
if self.dataset_conf.startswith("textvqa"):
self.dataset_path = os.path.join(self.dataset_path,
"textvqa_val.jsonl")
with open(conf_path, 'r', encoding='utf-8') as f:
content = f.read()
content = re.sub(r'path=.*', f'path="{self.dataset_path}",',
@@ -180,9 +183,13 @@ class AisbenchRunner:
def _get_result_performance(self):
result_dir = re.search(r'Performance Result files locate in (.*)',
self.result_line).group(1)[:-1]
result_csv_file = os.path.join(result_dir, "gsm8kdataset.csv")
result_json_file = os.path.join(result_dir, "gsm8kdataset.json")
dataset_type = self.dataset_conf.split('/')[0]
result_csv_file = os.path.join(result_dir,
f"{dataset_type}dataset.csv")
result_json_file = os.path.join(result_dir,
f"{dataset_type}dataset.json")
self.result_csv = pd.read_csv(result_csv_file)
print("Getting performance results from file: ", result_json_file)
with open(result_json_file, 'r', encoding='utf-8') as f:
self.result_json = json.load(f)

49
tools/send_mm_request.py Normal file
View File

@@ -0,0 +1,49 @@
import base64
import os
import requests
from modelscope import snapshot_download # type: ignore
mm_dir = snapshot_download("vllm-ascend/mm_request", repo_type='dataset')
image_path = os.path.join(mm_dir, "test_mm2.jpg")
with open(image_path, 'rb') as image_file:
image_data = base64.b64encode(image_file.read()).decode('utf-8')
data = {
"messages": [{
"role":
"user",
"content": [{
"type": "text",
"text": "What is the content of this image?"
}, {
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{image_data}"
}
}]
}],
"eos_token_id": [1, 106],
"pad_token_id":
0,
"top_k":
64,
"top_p":
0.95,
"max_tokens":
8192,
"stream":
False
}
headers = {'Accept': 'application/json', 'Content-Type': 'application/json'}
def send_image_request(model, server):
data["model"] = model
url = server.url_for("v1", "chat", "completions")
response = requests.post(url, headers=headers, json=data)
print("Status Code:", response.status_code)
response_json = response.json()
print("Response:", response_json)
assert response_json["choices"][0]["message"]["content"], "empty response"