[MODEL] Adding Support for Qwen3.5 Models#18489

Merged

mickqian merged 48 commits intosgl-project:mainfrom

zju-stu-lizheng:dev/qwen3.5

Feb 9, 2026

Commits on Jan 21, 2026

[qwen3_next vl] first push draft model file
瑀澈
committed

Commits on Jan 22, 2026

[qwen3_next vl] load moe, need to fix some bugs
瑀澈
committed
[mrope] fix for transformers v5:
瑀澈
committed
[qwen3_next vl] support moe version
瑀澈
committed
[qwen3_next vl] support dense model
瑀澈
committed

Commits on Jan 23, 2026

[lint] delete debug info
瑀澈
committed
[qwen3_next vl] support mtp model
瑀澈
committed

Commits on Jan 24, 2026

fix : add multimodel config for qwen3.5
瑀澈
committed
refactor(srt): Renamed modules and simplified the Qwen 3.5 model codebase
瑀澈
committed
fix: rope_theta and partial_rotary_factor factor
瑀澈
committed
[qwen3_5] add qwen3_5 dense and moe model
瑀澈
committed
fix(srt): import FusedMoE
瑀澈
committed

Commits on Jan 29, 2026

fix: offline preprocessed video
瑀澈
committed

Commits on Feb 4, 2026

1. fallback qwen3vl
瑀澈
committed
fix: remove visual parameter for Qwen3_5ForCausalLM
瑀澈
committed

Commits on Feb 5, 2026

fix import
瑀澈
committed
[qwen3_next vl] first push draft model file
瑀澈
committed
[qwen3_next vl] load moe, need to fix some bugs
瑀澈
committed
[mrope] fix for transformers v5:
瑀澈
committed
[qwen3_next vl] support moe version
瑀澈
committed
[qwen3_next vl] support dense model
瑀澈
committed
[lint] delete debug info
瑀澈
committed
[qwen3_next vl] support mtp model
瑀澈
committed
fix : add multimodel config for qwen3.5
瑀澈
committed
refactor(srt): Renamed modules and simplified the Qwen 3.5 model codebase
瑀澈
committed
fix: rope_theta and partial_rotary_factor factor
瑀澈
committed
[qwen3_5] add qwen3_5 dense and moe model
瑀澈
committed
fix(srt): import FusedMoE
瑀澈
committed
fix: offline preprocessed video
瑀澈
committed
1. fallback qwen3vl
瑀澈
committed
fix: remove visual parameter for Qwen3_5ForCausalLM
瑀澈
committed
fix import
瑀澈
committed
remove Qwen3NextVLConfig
瑀澈
committed
Merge branch 'yuche/qwen_next_vl' of http://gitlab.alibaba-inc.com/DamoAGI/sglang-internal into yuche/qwen_next_vl
瑀澈
committed
fix(logits_processor): clean up redundant code and merge conflict markers
瑀澈
committed
fix: support RadixLinearAttention
瑀澈
committed
fix: add target_extend_input_embeds for mtp(vl embedding)
瑀澈
committed

Commits on Feb 7, 2026

refactor(models): optimize convolution weight passing logic in Qwen3_5 model
瑀澈
committed
fix: add activation type
瑀澈
committed
fix: reduce_results for out_proj
瑀澈
committed

Commits on Feb 8, 2026

Commits on Feb 9, 2026

fix: indent for deepstack
瑀澈
committed
refactor(srt): Streamline redundant comments in Qwen3_5MultiTokenPredictor
瑀澈
committed
feat(moe): fix mamba cache support
瑀澈
committed
fix: mamba cache support Qwen3_5ForConditionalGeneration
瑀澈
committed
pre-commit
瑀澈
committed
fix qwen3.5 video preprocess
瑀澈
committed