[model] support new model - #10

Open

ltd0924 wants to merge 49 commits into

mainfrom

Commits on May 28, 2026

[ROCm][CI] Move workload from MI300 to MI325 (#43824)
AndreasKaratzas
authored
a9bc0ad
[Feature] Add support for timed trace replay in `vllm bench serve` to replay Moonshot and Alibaba workload traces (#39795)
animeshtrivedi
authored
bfb9ebc
[UX] Increase DP Coordinator startup timeout from 30s to 120s (#42343)
wzhao18
authored
f2caefe
[Model][Bugfix] Rename weight_mapper to hf_to_vllm_mapper in LlamaNemotronVL pooling models (#43581)

authored
4ec2817
[Bugfix][ROCm] Fix Accuracy Drop in Sparse Indexer on gfx950 (#43781)

authored
a583c84
[Bugfix] Fix HyperCLOVAX CI failure after upstream removed remote code (#43860)

khluu
and
claude
authored
61288b5
[CI] Auto-apply `rust` label to relevant PRs (#43866)
BugenZhao
authored
8e0580f
[Feature] Add structured output and effort support to Anthropic Messages API (#42396)
chaunceyjiang
authored
d692b89
Log dummy DP step in iteration details (#41406)
vadiklyutiy
authored
c1c4db8
[EC Connector] Add shutdown API to EC Connector. (#42423)
omerpaz95
authored
811d805
Fix `OlmoHybridForCausalLM` not initialising (#43846)

hmellor
and
mergify[bot]
authored
19af4e6
[BUGFIX] Multimodal benchmark with MistralTokenizer (#42965)
juliendenize
authored
02606b0
support step3.7
luotingdan
committed
ff8e2f4
Add Step3p5 MTP speculative decoding support
luotingdan
committed
4f6f07b
Clean up Step3p5 MTP runner wiring
luotingdan
committed
c1559bc
Fix Step3p5 MTP per-group slot_mapping for hybrid SWA
luotingdan
committed
cd3e007
Slice per-group attn metadata to unpadded shape in Step3p5 MTP
luotingdan
committed
a6a5d94
[Model] support step3.7 flash
luotingdan
committed
45e5a9a
update
luotingdan
committed
c99e8a5
add test registry
luotingdan
committed
ea7f396
fix render

Isotr0py
authored and
luotingdan
committed
2b148eb
fix: support Step3.7 ModelOpt NVFP4 loading

zidanehuang001
authored and
luotingdan
committed
51b948c
workaround for 3.7 FP8

jeejeelee
authored and
luotingdan
committed
1c50616
fix mtp3
luotingdan
committed
5f02c10
Restore tokenizer registry patch context
luotingdan
committed
28278e2
Revert test_mtp changes
luotingdan
committed
2b358ae
Revert tokenizer registry changes
luotingdan
committed
4b52a17
Revert vllm config change
luotingdan
committed
ec13752
Revert "Revert tokenizer registry changes"
luotingdan
committed
d74beb2
Use model type tokenizer cache condition
luotingdan
committed
01746e2
Fix Step3p5 MTP mypy typing
luotingdan
committed
6983152
fix format
luotingdan
committed
3660097
[Perf] Optimize moe permute by pre-allocate buffer, 9~14% kernel performance improvement (#43014)
yewentao256
authored
64e1218
set available false
luotingdan
committed
6f35039
[Perf][KDA] Fuse gate softplus, chunk-local cumsum, and RCP_LN2 scaling (#43667)

zexplorerhj
and
phyde682
authored
f3b2a81
Add token-offset based selective offload in OffloadConnector (#39983)

ruocco
and
orozery
authored
864990e
[Model Refactoring] Remove torch compile dependency in DSv4 (#43746)
WoosukKwon
authored
9957e4d
[Bugfix][ROCm] Resolve MoRI connector hangs at high concurrency (#40344)
simondanielsson
authored
552eb81
[CPU] Migrate cpu_awq into awq_marlin (#43841)
bigPYJ1151
authored
20d69d1
[Rust Frontend] Add `hy_v3` tool parser (#43872)
BugenZhao
authored
3a28223
[Rust Frontend] Reduce Gemma4 tool parser args scan complexity (#43850)
BugenZhao
authored
61a1e30
[rust] fix: aggregate `is_sleeping` and `reset_prefix_cache` across DP engines (#43429)

willamhou
and
claude
authored
577d693
[Bug] Fix `tests/distributed/test_elastic_ep.py - assert False` (#43813)
yewentao256
authored
be4062f
[Perf] Add do_not_specialize to Mamba SSD chunk kernels (#43803)

Majid-Taheri
and
Majid Taheri Andani
authored
c08ebeb
[Bugfix] Exclude Ray DP from #42585's deferred port allocation (#43864)
vadiklyutiy
authored
5d126dd
[KV Offload] Rename `SecondaryTierManager.get_finished()` to `get_finished_jobs()` (#43870)
ronensc
authored
4bfa0f2
[ROCm][Perf] Support N=5 in wvSplitK skinny GEMM kernels for speculative decoding (#40687)
mgehre-amd
authored
a9ec46d
[XPU][MoE] Add WNA16 oracle backend for GPTQ sym-int4 (xpu_fused_moe) (#41426)

authored
3207e76
Merge branch 'main' into feat/step3p7
ltd0924
authored
e4d7fa9