vlm: enable radix cache for qwen-vl models#5349

Merged

merrymercy merged 19 commits intosgl-project:mainfrom

mickqian:qwen-vl-radix-cache

Apr 24, 2025

Commits on Apr 24, 2025

move mrope to mm_processor
mickqian
committed
cleanups
mickqian
committed
process_mm_data_async signature
mickqian
committed
qwen2_5_vl test passed
mickqian
committed
qwen2vl test passed
mickqian
committed
fix minicpmo
mickqian
committed
remove audio_data param, move and merge mrope positions
mickqian
committed
tmp
mickqian
committed
change mrope_position back to tensor
mickqian
committed
mmmu prepare_samples
mickqian
committed
Refactor mrope position handling in ForwardBatch

JustinTong0323
authored and
mickqian
committed
fix: extend mrope positions

JustinTong0323
authored and
mickqian
committed
fix: extend mrope positions

JustinTong0323
authored and
mickqian
committed
skip ci: test_gme_qwen_models

JustinTong0323
authored and
mickqian
committed
fix gme test
mickqian
committed
put gme test back

JustinTong0323
authored and
mickqian
committed
gemma3 and mrope
mickqian
committed
bugfix
mickqian
committed
update test_openai_server test time
mickqian
committed