minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

fix: adapt dots_vlm for transformers v5 (#25829)

Browse Source
This commit is contained in:
hanwlax
2026-05-28 15:26:13 +08:00
committed by GitHub
parent 50e0b3b77f
commit 794fdd39ef
1 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/multimodal/processors/dots_vlm.py
+10 -2
View File
@@ -33,8 +33,16 @@ class DotsVLMImageProcessor(BaseMultimodalProcessor):
merge_size = vision_config.spatial_merge_size merge_size = vision_config.spatial_merge_size
self.IMAGE_FACTOR = patch_size * merge_size self.IMAGE_FACTOR = patch_size * merge_size
self.MIN_PIXELS = _processor.image_processor.min_pixels self.MIN_PIXELS = getattr(
self.MAX_PIXELS = _processor.image_processor.max_pixels _processor.image_processor,
"min_pixels",
getattr(_processor.image_processor, "size", {}).get("shortest_edge"),
)
self.MAX_PIXELS = getattr(
_processor.image_processor,
"max_pixels",
getattr(_processor.image_processor, "size", {}).get("longest_edge"),
)
self.MAX_RATIO = 200 self.MAX_RATIO = 200
self.mm_tokens = MultimodalSpecialTokens( self.mm_tokens = MultimodalSpecialTokens(
image_token=self.IMAGE_TOKEN, image_token=self.IMAGE_TOKEN,