EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix chat template handling for OpenAI serving (#8635)

Browse Source 
Signed-off-by: Xinyuan Tong <justinning0323@outlook.com>
Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
This commit is contained in:
Xinyuan Tong
2025-07-31 21:49:45 -07:00
committed by GitHub
parent 20b5563eda
commit 7e831efee8
7 changed files with 83 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
python/sglang/srt/multimodal/processors/gemma3n.py
View File

@@ -12,7 +12,6 @@
# limitations under the License.
# ==============================================================================
import re
from typing import Dict, List, Optional, Union
from sglang.srt.managers.multimodal_processor import (
@@ -38,14 +37,8 @@ class Gemma3nSGLangProcessor(SGLangBaseProcessor):
self.mm_tokens = MultimodalSpecialTokens(
image_token="<image_soft_token>",
image_token_id=hf_config.image_token_id,
image_token_regex=re.compile(
r"<start_of_image>(?:(?:<image_soft_token>)*<end_of_image>)?"
),
audio_token="<audio_soft_token>",
audio_token_id=hf_config.audio_token_id,
audio_token_regex=re.compile(
r"<start_of_audio>(?:(?:<audio_soft_token>)*<end_of_audio>)?"
),
).build(_processor)
async def process_mm_data_async(