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

[BugFix][EPD] adapt for qwen3.5-mtp & del duplicated logs (#24144)

Browse Source
This commit is contained in:
Zheng Wengang
2026-05-23 15:32:57 +08:00
committed by GitHub
parent 5964d30233
commit 8b9fb13c4a
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/multimodal/processors/qwen_vl.py
+2 -1
View File
@@ -24,6 +24,7 @@ from sglang.srt.models.qwen3_5 import (
Qwen3_5ForConditionalGeneration, Qwen3_5ForConditionalGeneration,
Qwen3_5MoeForConditionalGeneration, Qwen3_5MoeForConditionalGeneration,
) )
from sglang.srt.models.qwen3_5_mtp import Qwen3_5ForCausalLMMTP
from sglang.srt.models.qwen3_omni_moe import Qwen3OmniMoeForConditionalGeneration from sglang.srt.models.qwen3_omni_moe import Qwen3OmniMoeForConditionalGeneration
from sglang.srt.models.qwen3_vl import Qwen3VLForConditionalGeneration from sglang.srt.models.qwen3_vl import Qwen3VLForConditionalGeneration
from sglang.srt.models.qwen3_vl_moe import Qwen3VLMoeForConditionalGeneration from sglang.srt.models.qwen3_vl_moe import Qwen3VLMoeForConditionalGeneration
@@ -247,6 +248,7 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
Qwen3VLMoeForConditionalGeneration, Qwen3VLMoeForConditionalGeneration,
Qwen3_5ForConditionalGeneration, Qwen3_5ForConditionalGeneration,
Qwen3_5MoeForConditionalGeneration, Qwen3_5MoeForConditionalGeneration,
Qwen3_5ForCausalLMMTP,
InternS2PreviewForConditionalGeneration, InternS2PreviewForConditionalGeneration,
Qwen3OmniMoeForConditionalGeneration, Qwen3OmniMoeForConditionalGeneration,
] ]
@@ -475,7 +477,6 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
embedding_start : embedding_start + num_tokens embedding_start : embedding_start + num_tokens
] ]
consumed_per_modality[modality] = embedding_start + num_tokens consumed_per_modality[modality] = embedding_start + num_tokens
logger.info(f"Get embedding slice for {modality}, num_tokens={num_tokens}")
mm_items.append( mm_items.append(
MultimodalDataItem( MultimodalDataItem(
modality=modality, modality=modality,