feat(mem_cache): page-major (layer-major within a page) KV/state layout (#29533)

Co-authored-by: lch1475369 <lch1475369@gmail.com>
This commit is contained in:
Cheng Wan
2026-06-29 14:49:54 -07:00
committed by GitHub
co-authored by lch1475369
parent 6c018eb4d1
commit fc96edd297
25 changed files with 2159 additions and 128 deletions
+2
View File
@@ -66,6 +66,8 @@ DEFAULT_MODEL_NAME_FOR_TEST_MLA_NEXTN = "lmsys/sglang-ci-dsv3-test-NextN"
# Hybrid Mamba models
DEFAULT_HYBRID_MAMBA_MODEL_NAME_FOR_TEST = "Qwen/Qwen3-Next-80B-A3B-Instruct"
# Small GDN-hybrid (gated delta net) model that fits a single GPU
DEFAULT_HYBRID_GDN_SMALL_MODEL_NAME_FOR_TEST = "Qwen/Qwen3.5-4B"
# VL test models
DEFAULT_MODEL_NAME_FOR_TEST_VL_PP = "Qwen/Qwen3-VL-2B-Thinking"
DEFAULT_MODEL_NAME_FOR_TEST_GLM_41V_PP = "zai-org/GLM-4.1V-9B-Thinking"